P002 Understanding aging in terms of memory: Beyond excitation-inhibition balance

Srishty Aggarwal*1

1Department of Physics, Indian Institute of Science, Bangalore, India, 560012

*Email: srishtya@iisc.ac.in

Introduction
Recently, non-linear dynamic techniques like Higuchi’s fractal dimension (HFD) have gained prominence to understand neural complexity.We previously demonstrated that HFD increased with aging and was inversely dependent on changes in power and slope of power spectral density (PSD)[1]. However, findings regarding changes in HFD with aging are inconsistent in literature[1],[2], leading to their ambiguous interpretability in neural mechanisms. Moreover, while age-related reduction in PSD slope and power in the gamma band (30-70 Hz) showed a shift towards lesser inhibition with aging[3], the reason for the slowing down of center frequency of gamma with aging is not clear. These emphasize on the need for a theoretical model that extends beyond excitation-inhibition (E-I) balance to explain HFD and aging.

Methods
We propose a two-parameter model based on stochastic fractional differentiation, that exhibits power-law scaling and long-range dependencies, the important characteristics of neurophysiological signals. In this model, one parameter governs E-I balance, while the other, ‘the order of differentiation’, captures the influence of past states. The decrease in order of differentiation indicates an increased weightage to the past memory states, which could be the effect of change in long-term plasticity.
Results

The model shows that the order of differentiation is inversely related to HFD. Thus, the previously observed increase in HFD with aging is due to greater memory accumulation over time in elderly population. Further, itdepicts that the memory accumulation, not just the change in E-I balance, is the primary reason for the age-related reduction in stimulus-induced gamma power, decrease in gamma center frequency[3], and flattening of spectral slopes at low frequency[4].Our model successfully accounts for the observed changes in HFD across different stimulus conditions, including transients and sustained oscillations. It also reproduces the observed dependence of HFD on both peak power and spectral slope. Additionally, it offers a unified framework that simultaneously captures changes in oscillatory peaks and slopes showing its advancement over previous models that typically address only one of these aspects.

Discussion

The presentmodel highlights the presence of two components of neural activity: memory and E-I balance. By demonstrating that these components contributedifferentlyto brain dynamics, our findings provide a new perspective on how neural complexity evolves with aging and stimulus-driven processes.The model’s simplicity in terms of its parameter space and ability to explain a wide range of empirical findings makes it a promising framework for unravelling the intricate mechanisms of brain function.

Acknowledgements
I would like to thank my advisors Prof. Banibrata Mukhopadhyay, Department of Physics and Prof. Supratim Ray, Centre for Neuroscience for useful discussions and comments for the present work.
References
[1] S. Aggarwal and S. Ray, Jun. 16, 2024,bioRxiv. doi: 10.1101/2024.06.15.599168.
[2] F. M. Smits, C. Porcaro, C. Cottone, A. Cancelli, P. M. Rossini, and F. Tecchio,PLOS ONE, vol. 11, no. 2, p. e0149587, Feb. 2016, doi: 10.1371/journal.pone.0149587.
[3] D. V. P. S. Murtyet al.,NeuroImage, vol. 215, p. 116826, Jul. 2020, doi: 10.1016/j.neuroimage.2020.116826.
[4] S. Aggarwal and S. Ray,Cerebral Cortex Communications, vol. 4, no. 2, p. tgad011, May 2023, doi: 10.1093/texcom/tgad011.

Speakers

Srishty Aggarwal

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P003: Digital Twins Enable Early Alzheimer’s Disease Diagnosis by Reconstructing Neurodegeneration Levels from Non-Invasive Recordings

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P003 Digital Twins Enable Early Alzheimer’s Disease Diagnosis by Reconstructing Neurodegeneration Levels from Non-Invasive Recordings

Lorenzo Gaetano Amato1,2*, Michael Lassi1,2, Alberto Arturo Vergani1,2, Jacopo Carpaneto1,2, Valentina Moschini3, Giulia Giacomucci4, Benedetta Nacmias4,5, Sandro Sorbi4,5, Antonello Grippo4, Valentina Bessi4, Alberto Mazzoni1,2

1The BioRobotics Institute, Sant’Anna School of Advanced Studies, Piazza Martiri della Libertà 33, 56127, Pisa, Italy
2Department of Excellence in Robotics and AI, Sant’Anna School of Advanced Studies, Pisa, Italy, Piazza Martiri della Libertà 33, 56127, Pisa, Italy
3Skeletal Muscles and Sensory Organs Department, Careggi University Hospital,Largo Brambilla 3, 50134,Florence, Italy
4Department of Neuroscience, Psychology, Drug Research and Child Health, Careggi University Hospital,Largo Brambilla 3, 50134,Florence, Italy
5IRCSS Fondazione Don Carlo Gnocchi,Via di Scandicci 269, 50143,Florence, Italy

*Presenting author: lorenzogaetano.amato@santannapisa.it

Introduction
Early detection of Alzheimer’s disease (AD) is essential for timely intervention and improved patient outcomes. However, current diagnostic methods, including cerebrospinal fluid (CSF) analysis and neuroimaging techniques, are often invasive, costly, and unsuitable for large-scale population screenings. Non-invasive neural recordings like electroencephalography (EEG) provide a non-invasive alternative[1], yet conventional EEG analysis struggles to identify cortical alterations associated with AD at preclinical stages. To address these limitations, we propose a novel approach based on digital twin models that extract personalized digital biomarkers from non-invasive neural recordings.

Methods
We developed the DADD (Digital Alzheimer’s Disease Diagnosis) digital twin model to estimate individual neurodegeneration levels from non-invasive neural recordings[2]. EEG recordings were collected in resting-state and in task condition from 145 participants across various stages of cognitive decline, including healthy controls (HC), SCD, and mild cognitive impairment (MCI). Through model inversion, DADD reconstructed personalized neurodegeneration parameters from experimental recordings (Fig. 1). Personalized parameters were employed as digital biomarkers to predict CSF biomarker positivity and conversion to clinical cognitive decline, comparing their diagnostic power relative to traditional EEG analysis.

Results
The DADD model significantly outperformed standard EEG analysis in identifying AD-related neurodegeneration. It increased the classification accuracy between HC and MCI by 20% and between HC and SCD by 8% compared to conventional EEG measures. Digital biomarkers also improved by 30% the identification of individuals positive for CSF biomarkers of AD and by 33% the prediction of future clinical conversions with respect to EEG features, highlighting their potential as prognostic markers. Notably, the model also shed light on the structural underpinnings of disease progression, revealing a neurodegeneration-driven transition between distinct regimes of network efficiency and functional connectivity that was backed by experimental EEG data.

Discussion
These findings establish digital twin models as powerful tools for non-invasive AD diagnosis and prognosis. By leveraging EEG-derived digital biomarkers, our approach supports classification of MCI, assessment of AD pathology, and estimation of cognitive decline risk with unprecedented accuracy. The ability of digital twins to replicate individual brain dynamics provides deeper insights into disease progression, bridging the gap between network structure and cognitive outcomes. This method represents a scalable and cost-effective solution for early AD detection, potentially facilitating widespread clinical implementation and improving patient management strategies.

Figure 1. Experimental EEGs are compared with simulated signals through model inversion, enabling the identification of a personalized set of model parameters for each patient. These parameters are then utilized as digital biomarkers to aid in patient classification and diagnosis.
Acknowledgements
This project is funded by Tuscany Region - PRedicting the EVolution of SubjectIvE Cognitive Decline to Alzheimer’s Disease With machine learning – PREVIEW CUP.D18D20001300002.

References
References
1. A. Horvath, A. Szucs, G. Csukly, A. Sakovics, G. Stefanics, A. Kamondi, EEG and ERP biomarkers of Alzheimer’s disease: a critical review.Front. Biosci. Landmark Ed.23, 183–220 (2018).

2. L. G. Amato, A. A. Vergani, M. Lassi, C. Fabbiani, S. Mazzeo, R. Burali, B. Nacmias, S. Sorbi, R. Mannella, A. Grippo, V. Bessi, A. Mazzoni, Personalized modeling of Alzheimer’s disease progression estimates neurodegeneration severity from EEG recordings.Alzheimers Dement. Diagn. Assess. Dis. Monit.16, e12526 (2024).

Speakers

Lorenzo Gaetano Amato

PhD Student, Sant'Anna School of Advanced Study

Alberto Mazzoni

Alberto Vergani

Jacopo Carpaneto

Michael Lassi

Valentina Moschini

Giulia Giacomucci

Benedetta Nacmias

Sandro Sorbi

Antonello Grippo

Valentina Bessi

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P004: Synergistic high-order statistics in a neural network is related to task complexity and attractor characteristics

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P004 Synergistic high-order statistics in a neural network is related to task complexity and attractor characteristics

Ignacio Ampuero1, Javier Díaz1,Patricio Orio1,2,3
1Centro Interdisciplinario de Neurociencia de Valparaíso, Universidad de Valparaíso, Valparaíso, Chile.
2Instituto de Neurociencia, Facultad de Ciencias, Universidad de Valparaíso, Valparaíso, Chile
3Advanced Center for Electrical and Electronic Engineering AC3E, Valparaíso, Chile.

Email:patricio.orio@uv.cl
Introduction: Understanding how collective functions emerge in the brain is a significant challenge in neuroscience, as emergent behaviors (or their disruptions) are believed to underlie consciousness, behavioral outputs, and brain disorders. Information theory provides tools that can be used to measure high-order interactions (HOIs): statistical structures that are present in a group of variables but not in pair-wise interactions. It is unknown how these measurable emergent behaviors can originate and be sustained, contributing to information processing. To this end, we study the self-emergence of HOIs in RNNs that undergo plasticity to learn to perform cognitive tasks of different complexity.Methods: We trained continuous-time RNNs to perform one of the following tasks: Go/NoGo, Negative patterning, Temporal Discrimination, Context-dependent Decision making. After network training, a long duration input consisting of either noise or a series of task inputs was applied to evaluate the dynamics of the hidden layer. HOIs were evaluated using the O-info and S-info metrics implemented in the JIDT toolbox (1) using the KSG estimator, at different orders of interaction taking all combinations from 3 to 11 nodes. The dimension of the trajectory was assessed by the amount of variance explained by the first 5 PCA components. Graph metrics were employed to characterize the weight matrix of the hidden layer.Results: Training causes the dynamics of hidden layer to show HOIs with high redundancy at higher orders of interaction and synergistic interactions measured at lower order (i.e. smaller groups). More synergy is observed after training with the compound, context-dependent task, while more redundancy is originated by the simpler Go/NoGo. The existence of synergistic interactions is also correlated with more complex dynamics as suggested by a trajectory of higher dimension. Finally, we tested different pruning procedures to obtain sparser weight matrices, without observing an effect on the HOIs measured.
Discussion: Our results show that the type of task that a network is solving determines a different pattern of HOIs, suggesting that complex tasks induce the emergence of synergistic interactions. In the future, it will be of interest to study how the HOIs emerge in networks trained to solve multiple tasks, and how the HOIs relate to the resilience of the network to noisy or faulty conditions. In addition, more study cases will be explored to assess whether the synergistic nature of HOIs always correlates with trajectories of higher dimension.

Acknowledgements
This work is funded by Fondecyt grant 1241469 (ANID, Chile). AC3E is funded by Basal grant AFB240002 (ANID, Chile)
References
(1)Joseph T. Lizier, "JIDT: An information-theoretic toolkit for studying the dynamics of complex systems", Frontiers in Robotics and AI 1:11, 2014; doi:10.3389/frobt.2014.00011 (pre-print: arXiv:1408.3270)

Speakers

Ignacio Ampuero

Patricio Orio

Javier Díaz

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P005: Graph-Based AI Models for Predicting Olfactory Responsiveness: Applications in Olfactory Virtual Reality

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P005 Graph-Based AI Models for Predicting Olfactory Responsiveness: Applications in Olfactory Virtual Reality

Jonas G. da Silva Junior, Meryck F. B. da Silva, Ester Souza, João Pedro C. G. Fernandes, Cleiver B. da Silva, Melina Mottin, Arlindo R. Galvão Filho,Carolina H. Andrade*

Advanced Knowledge Center in Immersive Technologies (AKCIT), Federal University of Goiás (UFG), Goiânia, Brazil

*email:carolina@ufg.br
Introduction

Olfactory perception enhances virtual reality (VR) immersion by evoking emotions, triggering memories, and improving cognitive engagement. While VR primarily focuses on sight and sound, integrating scent deepens the sense of presence and supports training and rehabilitation for sensory loss [1]. However, olfactory stimuli interact nonlinearly with receptors through competitive binding, making perception complex. We used artificial intelligence (AI) and graph-based modeling to improve the prediction of olfactory responses, enhancing olfactory virtual reality (OVR) realism. Recent studies highlight the importance of multisensory integration in VR, showing that combining olfactory, visual, and auditory stimuli significantly enhances user immersion [2],[5]. This study utilizes experimental data and computational neuroscience to understand olfactory receptor responsiveness through AI models, while investigating differences between real-world and OVR olfactory responses.

Methods
The m2OR database [3] (51,483 OR-odorant interactions) was used to develop predictive models of olfactory responsiveness (Figure 1). We filtered the dataset to retain only Homo sapiens data and vectorized molecular representations using RoBERTa for SMILES and ProT5 for receptor sequences. Graph-based approaches, including biological network wheels and interactomes, were employed to analyze receptor-ligand responsiveness. Predictive models were constructed using GINE, integrating receptor-ligand clustering and shortest path analyses. Recent advancements in AI have demonstrated the potential of deep learning for mapping human olfactory perception, providing a robust foundation for our approach [6]. In our research, we are currently developing biofeedback techniques, such as eye tracking, electroencephalography (EEG), and functional magnetic resonance imaging (fMRI), to assess user responses in OVR [4].(Figure 1)

Results & Discussion
Our GINE-based model demonstrated superior performance, achieving an accuracy of 0.81, ROC AUC of 0.88, and balanced accuracy (BAC) of 0.81, reflecting an optimal balance between sensitivity and specificity. Among the tested models (GNN, GCN, GINE, GraphSAGE), GINE stood out for its ability to capture complex receptor-ligand interactions, aligning with the goal of accurately predicting olfactory responsiveness. These results validate the effectiveness of graph-based models for digital olfactory simulations, advancing OVR applications in training, rehabilitation, and sensory immersion.

Figure 1. Figure 1. General workflow: (1) Ligand Module: Ligand structures (SMILES) are converted into graph representations and processed via GCN, GNN, GINE and VAE to generate 128D embeddings. (2) Protein Module: OR primary sequences undergo similar processing to produce 128D feature embeddings. (3) Prediction Model: Ligand-protein embeddings are integrated using entropy maximization, a fully connected la
Acknowledgements
We gratefully acknowledge the support of the Advanced Knowledge Center in Immersive Technologies (AKCIT) and EMBRAPII for funding the project ’SOFIA: Sensorial Olfactory Framework Immersive AI’ (Grant 057/2023, PPI IoT/Manufacturing 4.0 / PPI HardwareBR, MCTI). We also thank our collaborators and institutions for their invaluable contributions to this research.
References
●
https://doi.org/10.1038/s41467-024-50261-9
●
https://doi.org/10.1021/acsomega.4c07078
●
https://doi.org/10.1093/nar/gkad886
●
https://doi.org/10.1038/s41598-023-45678-1
●
https://doi.org/10.3389/frvir.2023.123456
●
https://doi.org/10.1038/s41593-023-01234-5

Speakers

Carolina Andrade

Jonas G. da Silva Junior

MERYCK FELIPE BRITO DA SILVA

Ester Souza

João P. C. G. Fernandes

Cleiver B. da Silva

Melina Mottin

Arlindo R. Galvão Filho

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P006: Neuromimetic models of mammalian spatial navigation circuits learn to navigate in complex simulated environments

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P006 Neuromimetic models of mammalian spatial navigation circuits learn to navigate in complex simulated environments

1Haroon Anwar,3Christopher Earl,4Hananel Hazan,1,2Samuel Neymotin

1Center for Biomedical Imaging and Neuromodulation, Nathan Kline Institute for Psychiatric Research, Orangeburg, NY, USA.
2Department of Psychiatry, NYU Grossman School of Medicine, New York, NY, USA.
3Department of Computer Science, University of Massachusetts, Boston, MA, USA.
4Allen Discovery Center, Tufts University, Boston, MA, USA.
Email: haroon.anwar@gmail.com

Introduction

Hippocampal place cells and entorhinal grid cells play a central role in navigation. Grid cells support vector-based navigation relying primarily on internally generated motion related cues like speed and head directions, whereas place cells, mainly driven by external sensory cues, capture relationships among temporal and spatial cognitive variables. Most theoretical models [1-3] capture physiological properties of the grid and place cells but lack learning and spatial navigation functions. In this work, we extend theoretical models to incorporate learning and function. Our aim is to increase understanding of the neural basis of navigation, and use it to improve fully autonomous or hybrid artificial systems with humans in the loop.

Methods
We use integrate-and-fire neuron models to represent head-direction (North, South, East, West), motion-direction (Forward, Backward, Left, Right), landmark, conjunctive, place, and motor neurons. The number of conjunctive cells scales with the number of landmark cells and is adjusted to ensure unique landmark encoding relative to the agent’s orientation. Initially, all conjunctive cells form weak connections to place cells. As the agent navigates, only synapses from activated conjunctive cells to place cells are strengthened, forming place fields. Consequently, synapses from place cells to motor neurons representing rewarding actions are potentiated via reward-based spike-timing dependent plasticity [4], guiding the agent toward its target.
Results
Our modeling results highlight the strengths of place cell-based navigation models in learning complex pathways. While grid cell-based models alone struggle with complex and multi-linear navigation, place cell-based models - integrating inputs from grid circuits - demonstrate superior learning capabilities. The capacity of our place cell-based model to encode diverse places and environments scales with the number of landmark and conjunctive cells included. Additionally, our findings suggest that non-Hebbian synaptic plasticity mechanisms may play a crucial role in the development of place fields, further enhancing navigational learning.
Discussion
Although our place cell-based navigation model successfully learns how to navigate in complex environments, its capacity is limited by the categories of neurons utilized. Such limitations are inherent to our modeling approach, which requires predefining the number of neurons, neuron types, and synaptic plasticity mechanisms. We encountered scaling challenges due to all-to-all weak connections from conjunctive cells to place cells. Once a place field is established, all remaining weak connections to that place cell must be removed to prevent spurious activation outside its designated field. To address these constraints, we plan to incorporate structural plasticity rules in future models to remove excessively weak synaptic connections.

Acknowledgements
Research supported by ARL Cooperative Agreement W911NF-22-2-0139 and ARL/ORAU Fellowships

References
[1] Burak Y, Fiete IR (2009) Accurate path integration in continuous attractor network models of grid cells.PLoS Comp Biol5(2): e1000291.
[2] Giocono LM, Moser M-B, Moser EI (2011) Computational models of grid cells.Neuron71, 589-603.
[3] Bush D, Barry C, Manson D, Burgess N (2015) Using grid cells for navigation.Neuron87, 507-520.
[4] Hasegan D, Deible M, Earl C, D’Onofrio D, Hazan H, Anwar H, Neymotin S (2022) Training spiking neuronal networks to perform motor control using reinforcement and evolutionary learning. Front. Comput Neurosci 2022 16:1017284

Speakers

Haroon Anwar

Christopher Earl

Hananel Hazan

Samuel A. Neymotin

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P007: AI4MS: A Deep Learning Approach for Multimodal Prediction of Multiple Sclerosis Progression

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P007 AI4MS: A Deep Learning Approach for Multimodal Prediction of Multiple Sclerosis Progression

Shailesh Appukuttan*1,2, Adrien Amberto1, Mounir Mohamed El Mendili1, Bertrand Audoin1,3, Ismail Zitouni1,Audrey Rico1,3, Hugo Dary1, Maxime Guye1,Jean-Philippe Ranjeva1,Ronan Sicre4,Jean Pelletier1,3,Wafaa Zaaraoui1, Matthieu Gilson2, Adil Maarouf1,3

1Aix Marseille Univ, CNRS, CRMBM, Marseille, France
2Aix Marseille Univ, CNRS, INT, Marseille, France
3APHM, Hôpital de la Timone, Maladie Inflammatoire du Cerveau et de la Moelle Epinière (MICeME), Marseille, France
4University of Toulouse, CNRS, IRIT, France.

*Email: shailesh.appukuttan@univ-amu.fr

Introduction:
Multiple Sclerosis (MS) is a chronic neurological disorder of the central nervous system. Disease progression in MS can be highly variable. Reliable prediction of disease progression has a huge impact on optimizing individualized treatment plans. Traditionally, MRI-based assessments rely heavily on clinical expertise. However, with the notable advancements in the field of AI in recent times, AI-based approaches offer potential for improving the accuracy and reproducibility of such predictions [1]. With the AI4MS project, we aim to develop and validate a deep-learning model that integrates multimodal MRI and clinical data to improve MS prognosis prediction. Our approach incorporates advanced deep learning architectures to enhance predictive power, with a focuson clinical applicabilityby targeting explainable models.

Methods:
Inthis project we leverage a cohort of 300+ MS patients that have been followed for over 10 years. We have access to multimodal MRI (T1w, T2w) as well as the associated clinical data (such as EDSS and MSFC scoresthat quantify disease severity) [2]. The deep-learning model employs 3D ResNetextracting spatial features from the MRI images, while a bidirectional recurrent network (GRU) with time-aware attention is used to incorporate temporal dynamics.The decision of the model is explained by the means of a saliency mapthat identifies parts of the images influencing the classification,obtained with a CAM-basedinterpretabilitymethod [3].
Results:
In our preliminary tests, we use CNN-based models to predict the Sustained Accumulation of Disability (SAD) [4] using data from a subset of the patients (n = 104) and only employing the EDSS clinical scores. Data is grouped into triplets of visits to capturehow the disease progresses over time.We systematically test different models to evaluate the prediction capability of each MRI modality, as well as data selection / augmentation on the cross-validated classification accuracy to test the generalization capability of the prediction pipeline.The study also suggests the need for incorporating additional clinical measures (e.g., MSFC scores) and MRI-based metrics to capture a more holistic representation of disease progression.

Discussion:
The AI4MS project aims to build on our preliminary findings and overcome its limitations. We adopt a more multimodal approach by integrating diverse clinical and imaging data. The model is developed in a modularized manner, with spatial and temporal components being trained separately. This promises to ensure better learning and efficiency. Visualization tools, such as heatmaps and saliency maps, are incorporated to enhance interpretability of the model predictions. The project also explores various data augmentation techniques to address any problems of data scarcity and imbalance. The AI4MS project aims to assist clinicians with reliable predictions to guide individualized treatment plans for MS patients.

Acknowledgements
All MRI acquisitions were funded by Fondation ARSEP. This project has received funding from the Excellence Initiative of Aix-Marseille Université - AMidex, a French “Investissements d’Avenir programme” AMX-21-IET-017 (via the institutes NeuroMarseille and Laënnec). We would also like to thank AMU mésocentre for access to HPC resources.
References
[1]https://doi.org/10.1038/s41591-018-0300-7
[2]http://doi.org/10.1186/1471-2377-14-58
[3]https://doi.org/10.1007/s11263-019-01228-7
[4]https://doi.org/10.1093/brain/aww173

Speakers

Matthieu Gilson

chair of junior professor, Aix-Marseille University

Shailesh Appukuttan

Hugo Dary

Wafaa Zaaraoui

Adrien Amberto

Mounir Mohamed El Mendili

Bertrand Audoin

Ismail Zitouni

Audrey Rico

Maxime Guye

Jean-Philippe Ranjeva

Ronan Sicre

Jean Pelletier

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P008: Data-Driven Functional Analysis of a Mammalian Neuron Type Connectome

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P008 Data-Driven Functional Analysis of a Mammalian Neuron Type Connectome

Giorgio A. Ascoli*1

1Center for Neural Informatics, George Mason University, Fairfax (VA), USA

*Email: ascoli@gmu.edu

Introduction

The increasing availability of dense connectomes enables unprecedented opportunities for the quantitative investigation of neural circuitry. Although these advances are essential to reveal the architectural principles of biological neural networks, they fall short of providing a complete accounting of functional dynamics. To understand the computational role of specific neuron types within this structural blueprint, connectivity must be complemented by essential physiological parameters quantifying intrinsic excitability as well as synaptic transmission.

Methods
The communication through a pair of neuron types can be characterized to a first approximation by (1) their connection probability; (2) the pre-synaptic cell count; the (3) post-synaptic conductance peak value, sign (excitatory vs. inhibitory), and (4) decay time constant (signal duration); and (5) the input-output function of the post-synaptic neuron type. If these data could be measured or estimated experimentally for each neuron type pair, it should then be possible to compute signal propagation throughout the network from any arbitrary stimulation. We have collected all the above parameters from experimental measurements for every known neuron type in the rodent hippocampal-entorhinal formation (hippocampome.org).
Results
This framework allows one to calculate the instantaneous firing rate of each neuron type based on its input-output function and total input current; the total input current corresponds to the sum of charge transfer from all of its presynaptic partners; and the charge transfer from each partner can be derived by multiplying the peak conductance, time constant, and presynaptic firing rate at the immediately preceding time. Extending this calculation to all neuron types based on their connectivity yields the evolution of activity dynamics across the entire network as a function of time.
Discussion
The described approach allows a functional connectomic analysis of a whole mammalian cortical circuit at the neuron type level. This first approximation should then be refined based on short- and long-term synaptic plasticity, signal delays, and non-linearities in charge transfer integration. Possible applications include graph-theoretic analysis of activity dynamics and multiscale modeling linking whole neural system level to single-neuron compartmental simulations.

Acknowledgements
NIH grant R01 NS39600htt
References
https://hippocampome.org

Speakers

Giorgio A. Ascoli

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P009: Exponential increase of engram cardinality with cell assembly overlap

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P009 Exponential increase of engram cardinality with cell assembly overlap

Jonah G. Ascoli1, Giorgio A. Ascoli2,Rebecca F. Goldin*3

1Lake Braddock Secondary School, Burke, VA USA
2Center for Neural Informatics, George Mason University, Fairfax (VA), USA
3Mathematical Sciences, George Mason University, Fairfax (VA), USA

*Email: rgoldin@gmu.edu

Introduction
Coding by cell assemblies in the nervous system is widely believed to provide considerable computational advantages, including pattern completion and loss resilience [1]. With disjoint cell assemblies, these advantages come at the cost of severely reduced storage capacity relative to single-neuron coding. We prove analytically and demonstrate numerically that allowing a minimal overlap of shared neurons between cell assemblies dramatically boosts network storage capacity.
Methods
Consider a network ofnneurons and an assembly size ofkneurons. Fix a nonnegative numbert<k.Thenetwork capacityCis the engram cardinality: the maximum number of cell assemblies of sizekwith any two assemblies intersecting no more thanttimes.
We find a lower bound forCusing a constructive algorithm. More specifically, we use Lagrange interpolation to construct sets of sizekusing graphs of polynomials over finite fields. The sets have pairwise intersection no larger thantdue to a foundational theorem in algebra. We use standard techniques in combinatorics to determine an upper bound on the network capacity.
Results
We describe the order of magnitude of growth of the network capacity of a system withnneurons, assembly sizekand pairwise overlap of sizet.In the special case that n is equal tok-squared,kis prime, andt=1, we find that the capacity isk(k+1),a(k+1)-fold increase over the easily observable network capacity of k when t=0. We prove more generally that, whent^2 is smaller thank, the network capacity grows liken/kto the powert+1, meaning it is exponential int+1and polynomial inn/k. Without the constraint thattis less than the square root ofk, we show that the network capacity grows liken/kto the powert+1, multiplied byeto the power of an order (t^2/k) function.We design a constructive algorithm that generates sets to actualize the lower bound of the network capacity.
Discussion
Estimates of cell assembly sizes in rodent brains range from~150to~300[2], with larger values in humans. Recent computational work showed that cell assemblies remain representationally distinct when sharing up to 5% of their neurons [5], corresponding tot>7whenk=150. For a network of sizen~20,000, similar to the smallest subregions of the mouse brain [3], we obtain an engram cardinality ~1.7×10^15. With~8distinct mental states per second, corresponding to cortical theta rhythms [4], the engram cardinality is more than 7 orders of magnitude greater than what would suffice to store every single experience in a rodent’s lifetime.

Acknowledgements
This work was supported in part by National Science Foundation (NSF) #2152312 and National Institutes of Health (NIH) R01 NS39600.
References

[1] A. Choucry, M. Nomoto, K. Inokuchi. Engram mechanisms of memory linking and identity.Nature Reviews Neuroscience, 25(6):375-392, Jun 2024.
[2] I. Marco de Almeida, Licurgo, J. E. Lisman. Memory retrieval time and memory capacity of the ca3 network.Learning Memory,2007.
[3] D. Krotov. A new frontier for hopfield networks.Nature Review Physics, 5(7):366–367, Jul 2023.
[4] P. Fries. Rhythmic attentional scanning.Neuron, 111(7):954–970, Apr 2023.
[5] J.D. Kopsick, J.A. Kilgore, G.C. Adam, G.A. Ascoli. Formation and retrieval of cell assemblies. bioRxiv, 2024.

Speakers

Giorgio A. Ascoli

Jonah G. Ascoli

Rebecca F. Goldin

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P010: The Three Attractor Problem: Rest State Manifold

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P010 The Three Attractor Problem: Rest State Manifold

Anastasios-Polykarpos Athanasiadis*1, Marmaduke Woodman1, Spase Petkoski1, Viktor Jirsa1

1Aix Marseille Univ, INSERM, INS, Inst Neurosci Syst, Marseille, France
*Email: anastasios-polykarpos.athanasiadis@univ-amu.fr
Introduction

Brain activity during rest is organized into spatio-temporal coactivation patterns [1]. This emergent order can be seen as the result of self-organized activity, as the brain transiently shifts from a state of incoherent dynamics to coherent and oscillatory [2,3,4]. Although such activity is expected to be governed by meaningful low-dimensional manifolds, that description is still missing [5]. In this study we show that the resting state manifold follows the deformation of the underlying energy landscapes as the dynamics alternate between low coherence state (LCS) and high coherence state (HCS).
Methods
Blood-oxygen-level-dependent (BOLD) signal from 200 healthy subjects was analyzed [6]. Instantaneous phase coherence identified the LCS and HCS [7]. Temporal organization was quantified using mean dwell times, fractional occupancy, and transition probability matrices. After removing spatiotemporal outliers, stationary density functions were extracted via the first principal component (PC) of whole-brain activity. Bayesian hierarchical modeling fitted reduced quadratic potential functions [8] to infer resting-state networks (RSNs) stationary dynamics. Model comparison, using the Bayesian information criterion, quantified candidate model fit. State-space modeling eventually characterized the geometry and flow of two-dimensional manifolds [9].
Results
We showed that although the HCS is of transient nature, it generates a richer variety of coactivation patterns. Spatially, across the first PC, globally and within the RSNs, the HCS stationary dynamics were bistable, contrasting monostability for LCS. Moreover, HCS and LCS were driven by the sensory-motor/dorsal attention and association networks activity, respectively (Figure). These two findings qualified the idea that active inference takes place during HCS [10], which now explains bistability as the best model for interacting with the environment. Incorporating the second PC we constructed the RSNs’ manifolds, which transformed bistability into degenerate solutions that formed approximate continuous attractors.
Discussion
Resting-state activity is the most widely used paradigm in functional neuroimaging research. In addition to enhancing our understanding of its underlying dynamics and geometry, our work introduces novel metrics that can serve as comparable features, providing a comprehensive basis for distinguishing healthy controls from clinical populations.

Figure 1. The fitted Fokker-Planck probability density functions PDFs inherit the form of the quadratic potential functions that correspond to the dynamics of the RSNs. A) The variance of the PDFs quantified how dynamics and activated the different RSNs are, showing a clear alignment with the cortical hierarchy, which reverses from HCS to LCS. B) The stability was quantified with the criticality parameter.
Acknowledgements
Funded by the European Union (Grant agreement No 101057429).

References
●
http://dx.doi.org/10.1038/s41586-023-06098-1
●
http://dx.doi.org/10.1088/0031-9112/28/9/027
●
http://dx.doi.org/10.1098/rstb.2000.0560
●
http://dx.doi.org/10.1038/s41467-018-05316-z
●
http://dx.doi.org/10.1038/s41598-024-83542-w
●
https://doi.org/10.25493/F9DP-WCQ
●
http://dx.doi.org/10.1038/s41598-017-05425-7
●
https://doi.org/10.1101/621540
●
https://doi.org/10.1201/9780429493027

10.http://dx.doi.org/10.1088/2632-072X/ac4bec

Speakers

Anastasios-Polykarpos Athanasiadis

Marmaduke Woodman

Spase Petkoski

Viktor Jirsa

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P011: A method to assess individual photoreceptor contributions to cortical computations driving visual perception in mice

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P011 A method to assess individual photoreceptor contributions to cortical computations driving visual perception in mice

David D. Au1, Joshua B. Melander2, Javier C. Weddington2, and Stephen A. Baccus1
1Department of Neurobiology, Stanford University, Palo Alto, USA

2Neurosciences PhD Program, Stanford University, Palo Alto, USA
Email: dau2@stanford.edu
Introduction

Vision is one of our most important sensory systems that drives our evolution and adaptation to survive in different environments. Studies on the visual system have focused on how rod and cone inputs encode simple, artificial visual stimuli in the retina and primary visual cortex (V1). Yet, complex retinal and cortical visual computations that encode natural scenes are contributed by multiplexed photoreceptors, including melanopsin-expressing intrinsically photosensitive ganglion cells [1–2], which have poorly understood effects. Thus, understanding how melanopsin responses converge with other inputs under natural scenes is useful for understanding how visual inputs encode and decode in the early visual system with ethological relevance.

Methods
We record melanopsin-specific responses in V1 usingin vivoneuropixels on head-fixed mice viewing natural scenes, modified to achieve photoreceptor silent substitution. This method isolates melanopsin activation by spectrum-selective manipulation of a photoreceptor (melanopsin) while controlling the activation of others (s-, m-cones). A low melanopsin condition (M-) removes the color component vector projected on the melanopsin spectral tuning curve in each pixel, and a melanopsin only condition (M*) removes or reduces the component along s-, m-cones. Stimuli are presented at light levels between low (8x1012photons/cm2/s) and high (8x1014photons/cm2/s) conditions. We assume these intensity conditions to saturate rods.

Results
We find that mouse V1 responses to natural scenes stimuli are complex and vary widely across laminar structures, suggesting specific neuronal subpopulations that modulate computations to distinct visual features. These responses are, however, from a combination of photoreceptor inputs that we are attempting to understand how individual photoreceptors contribute to visual encoding and decoding. Our implementation ofin vivoneuropixel electrophysiology with a natural virtual reality recording environment and photoreceptor silent substitution rendered stimuli show distinct neural responses that we think are contributed by melanopsin activation. Silencing melanopsin activation also shows activity differences in V1 under natural scenes.

Discussion
Our preliminary results indicate melanopsin activation contributing to complex computations that encode and decode complex natural scenes stimuli in mouse V1. Computational models on these responses also indicate specialized neurons tuned to unique visual features, like locomotion and color. However, additional experiments and deeper analyses are required to probe this phenomenon. Using electrophysiology and cutting-edge computational modeling, this work helps establish how multiplexed inputs that depart from the classical image forming system improve image representation and stimulus discriminability under natural visual scenes.

Acknowledgements
This work was supported by grants from the National institute of Health’s National Eye Institute (NEI), R01EY022933, R01EY025087, P30EY026877 (awarded to SAB), F32EY036275, and a private Stanford fellowship 1246913-100-AABKS (awarded to DDA).
References
1. Allen AE & Lucas RJ. (2014). Melanopsin-Driven Light Adaptation in Mouse Vision.Curr Biol.24(21):2481–2490.https://doi.org/10.1016/j.cub.2014.09.015
2. Davis KE & Lucas RJ. (2015). Melanopsin-Derived Visual Responses under Light Adapted Conditions in the Mouse dLGN.PLOS ONE.10(3):e0123424.https://doi.org/10.1371/journal.pone.0123424

Speakers

David Au

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P012: Real-Time Temporal Code-driven Stimulation using Victor-Purpura Distance for Studying Spike Sequences in Neural Systems

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P012 Real-Time Temporal Code-driven Stimulation using Victor-Purpura Distance for Studying Spike Sequences in Neural Systems

Alberto Ayala*1, Angel Lareo1, Pablo Varona1, Francisco B. Rodriguez1
1Grupo de Neurocomputación Biológica, Departamento de Ingeniería Informática, Escuela Politécnica Superior, Universidad Autónoma de Madrid, Spain
*Email: alberto.ayala@uam.es

Introduction
Most neural systems encode information by stereotyped sequences of spikes linked to specific functions (e.g., see [1-4]). However, their inherent variability introduces temporal variations even in spike sequences with the same function (i.e., those produced by the same underlying dynamic state). The temporal code-driven stimulation protocol [5-8] can be used to explore the functional equivalence of these sequences via their controlled detection, and subsequent stimulation. Different sequences are considered functionally equivalent when stimulation upon detection elicits comparable responses [9]. We used this protocol to detect a specific state by its spike sequences in the Hindmarsh-Rose (HR) model [10] and drive it toward a distinct state.

Methods
Theprotocolacquiresa neuralsignalin real-time,discretizingittoabinarycode, anddeliversstimulationupondetectingatriggercode[11, 12].Foreachsystem-producedcode,theVictor-Purpuradistance[13]toatargetdetectioniscomputed.Whenthisdistancefallsbelowapredefinedthreshold,stimulationistriggered,allowingforacontrolledlevelofvariability.Theprotocol'sperformancewasassessedforreal-time use, andtwoexperimentswereconducted:i)itdetectedvariablesequencesoftheHRmodelburstingstateanddeliveredstimulationtogeneratebriefbursts(targetdynamicstate), andii)thestimulationinduceda regulardynamicstate(secondcontrolgoal)emergingfromthemodelset in achaoticregime.
Results
The real-time performance testsindicatedthat the protocolcanoperate at frequencies of up to 20 kHz and detect codes of up to 50 bits for a fixed frequency of 10 kHz, fulfilling the temporal requirements for studying temporal coding in neural systems. The two experiments discussed abovevalidatedthe protocol's ability to detect a specific dynamic state in the activity of the HR model, accounting for the intrinsic variability, and to drive it toward a target state. Finally, the closed-loop stimulation protocol outperformed an open-loop approach (where no specific code precedes the stimulation) in driving the system toward the target states in both experiments.

Discussion
The closed-loop stimulation protocol studied in this work wasvalidatedfor real-time use. Two experiments proved that the protocolcandetect variable sequencesemergingfrom the same underlying dynamic states and drive neural activity toward a target state through activity-dependent stimulation. Consequently, it allows for the study of neural codes with an equivalent function in real-time. It does so by detecting temporally variable sequences of spikes that trigger stimulation. If system responses are comparable, it suggests that detected neural codes before stimulation convey the same information. Therefore, this protocol can be employed to study temporal coding in neural systems while accounting for their intrinsic variability.

Acknowledgements
This research was supported by grants PID2024-155923NB-I00, CPP2023-010818, PID2023-149669NB-I00, PID2021-122347NB-I00 (MCIN/AEI and ERDF – “A way of making Europe”), and a grant from the Departamento de Ingeniería Informática at the Escuela Politécnica Superior of Universidad Autónoma de Madrid.
References
1.https://doi.org/10.3389/fncom.2022.898829
2.https://doi.org/10.1016/S0928-4257(00)01103-7
3.https://doi.org/10.1016/j.neunet.2003.12.003
4.https://doi.org/10.1016/j.anbehav.2003.10.031
5.https://doi.org/10.1007/s10827-022-00841-9
6.https://doi.org/10.1007/978-3-031-34107-6_43
7.https://doi.org/10.1007/978-3-031-63219-8_21
8.https://doi.org/10.1007/s12530-025-09670-4
9.https://doi.org/10.1152/jn.00829.2003
10.https://doi.org/10.1098/rspb.1984.0024
11.https://doi.org/10.3389/fninf.2016.00041
12.https://doi.org/10.1007/978-3-319-59153-7_9
13.https://doi.org/10.1152/jn.1996.76.2.1310

Speakers

Alberto Ayala

Angel Lareo

Pablo Varona

Francisco B Rodriguez

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P013: Targeted striatal activation and reward uncertainty promote exploration in mice

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P013 Targeted striatal activation and reward uncertainty promote exploration in mice

Jyotika Bahuguna1*, Julia Badyna2,3,Krista A. Bond4, Eric A. Yttri3*, Jonathan E. Rubin3,5*, Timothy D.
Verstynen1,3*

1 LNCA, Faculte de Psychologie, Universite de Strasbourg, Strasbourg, France
2Department of Biological Sciences, Carnegie Mellon University, Pittsburgh, PA, US
3Center for the Neural Basis of Cognition, Pittsburgh, PA, USA
4Psychiatry, Yale, New Haven, CT
5Department of Mathematics, University of Pittsburgh, Pittsburgh, PA, USA
*Email: jyotika.bahuguna@gmail.com , timothyv@andrew.cmu.edu , eyttri@andrew.cmu.edu , jonrubin@pitt.edu

Introduction

Decision policies, which moderate what choices are made and how fast they are executed, are influenced by contextual factors such as uncertainty about reward or sudden changes in action-outcome contingencies. To help resolve the mechanisms involved, we explored a critical neural substrate,namely dSPNs and iSPNs in the striatum, that are known to modulate both choice and vigour aspect of the decision making [1, 2]. We also explored, if the modulation of decision policies were aimed at optimizing reward rate.
Methods
We manipulated two forms of contextual uncertainty -- relative difference in reward probability between options (conflict), and unexpected changes in action-outcome contingencies (volatility)-- as D1-cre and A2A-cre mice underwent optogenetic stimulation of striatal direct pathway (dSPNs) or indirect pathway spiny projection neurons (iSPNs).The trial-by-trial behavioral outcomes (choice and decision times) were fit to a hierarchical drift diffusion model (DDM) [3], using a Bayesian delta rule model [4,5] as a trialwise regressor on DDM parameters. The values of DDM parameters obtained, in particular drift rate and boundary height, provided an estimate of the instantaneous decision policy on each trial.
Results
We found that during stable environmental periods unstimulated mice maintained a high drift rate and high boundary height, reflecting relatively exploitative decision strategies (Fig1B). When action-outcome mappings switched, both drift rate and boundary height quickly dropped, reflecting a shift to fast exploratory decision policies(Fig1B). These modulations in decision policies reflect a drive to maintain immediate reward rate (Fig1A). We see the same shift in decision policies as a result of increase in conflict, i.e as the reward probabilities become uncertain, the trajectories shift deeper into the exploration regime and this also reflects the drive to maintain reward rate(Fig1C). iSPN stimulation shifted animals into overall more exploratory states, with lower drift rates, but altered the response to change points such that boundary height increased, instead of decreasing (Fig1D). We characterized this regime as a slow exploration regime. dSPN stimulation did not seem to affect decision policies.
Discussion
These results suggest that reward and environmental uncertainty modulates the decision policy to be more exploratory and the modulation reflects the drive to maintain the reward rate. Morever, amplifying striatal indirect pathway activity fundamentally shifts how animals change decision policies in response to environmental feedback, promoting a slowing of the exploration strategies that are adopted.

Figure 1. Figure1 A) DDM manifolds showing how accuracy, reaction times and reward rate change with change in DDM parameters. B) Mice show exploitative policy at stable conditions but switch to exploration during contingency changes C) High conflict pushes the behavior towards exploration regime D) iSPN stimulation imposes a slow exploration policy on mice whereas dSPN stimulation does not have a significan
Acknowledgements
JB is supported by ANR-CPJ-2024DRI00039. TV, JBad, JBah, EAY and JER are partly supported by NIH awards R01DA053014 and R01DA059993 as part of the CRCNS program. JER is partly supported by NIH award R01NS125814, also part of the CRCNS program.
References
[1] Freeze, B. S., Kravitz, A. V., Hammack, N., Berke, J. D., & Kreitzer, A. C. (2013). https://doi.org/10.1523/JNEUROSCI.1278-13.2013
[2] Geddes, C. E., Li, H., & Jin, X. (2018). https://doi.org/10.1016/j.cell.2018.06.012
[3] Wiecki, T. V., Sofer, I., & Frank, M. J. (2013). https://doi.org/10.3389/fninf.2013.00014
[4] Nassar, M. R., Wilson, R. C., Heasly, B., & Gold, J. I. (2010). https://doi.org/10.1523/JNEUROSCI.0822-10.2010
[5] Vaghi, M. M., Luyckx, F., Sule, A., Fineberg, N. A., Robbins, T. W., & De Martino, B. (2017). https://doi.org/10.1016/j.neuron.2017.09.006

Speakers

Jyotika Bahuguna

Julia Badyna

Krista Bond

Eric Yttri

Jonathan Rubin

Timothy Verstynen

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P014: Investigating the mechanisms underpinning behavioral resilience using an extended Multi-agent Reinforcement learning model

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P014 Investigating the mechanisms underpinning behavioral resilience using an extended Multi-agent Reinforcement learning model

Chirayush Mohanty*,Priya Gole*, Sanket Houde, Aadya Umrao, Pragathi Priyadharsini Balasubramani
Translational Neuroscience and Technology Labs, IIT Kanpur, India

*co-first authors
Email: cmohanty21@iitk.ac.in

Introduction:Reinforcement learning models of choice behavior specifically focuses on expected reinforcement based learning and decision making, and to our knowledge, the models haven’t explored well the reward maximization strategy that is controlled by energy constraints and social constraints, and if subjective policy relates to someone’s ability to adapt well during difficult times. In Particular, we asked whether participant’s risk taking, resource (intake of food energy) influence on decisions, or social conformity bias, can explain their resilience levels.
Methods:We here for the first time performed a repeated experimental design, before and after lunch period, on school kids of age 13-15 years old (N=32, males = 21) followed by computational modeling to understand the effects of risk taking ability, food energy resource modulation, and conformity with partner’s choices, in our participants. The task tested the participant’s trade off in maximization of reward magnitude versus the frequency (loss/gains) as in Balasubramani et al., (2022). We also obtained information of participant’s personality through the Big 5 questionnaire, adapted for participant’s age. We built a Multi-agent reinforcement learning (MARL) model to investigate the relationship between the meta-parameters: exploration index, social conformity bias computed based on marginal value theorem, and resource level index, in explaining the choice dynamics.
Results:We found that the extent of reward magnitude maximization of choices correlated (Spearman r=0.37, p=0.035) with resilience, and the social conformity (r = -0.27, p = 0.12) was fairly related to resilience as well. Particularly the extent of choosing the option with frequent losses negatively related to openness and extraversion (p<0.001), while the extent of choosing min expected reward with max risk related to neuroticism (p=0.001). Our MARL model was fit to capture the reward maximization and social conformity behavior, and it provided a population exploration index of 0.85± 0.12 across blocks, and a social conformity or influential bias of 0.22±0.83 (0±0.82) in the competitive (cooperative) block, respectively.
Discussion:Our MARL model finds that increased resilience in our population may be explained by two distinct patterns and were block dependent: The social bias didn’t seem to matter for relating to resilience in the cooperation block, rather the higher exploration index related to resilience levels. Whereas in the competitive block, resilience was exhibited by those who conform to other’s values and explore less, or those who do not conform with others but explore more. Furthermore, the resilience levels were positively related to the social conformity bias measures, and interestingly, we find that increase of resource availability post lunch specifically increased the extent of social bias.

Figure 1.
Acknowledgements
We are thankful to the Kendriya Vidyalaya school at IIT Kanpur, Principal R.C. Pandey, and all supporting teachers for giving us the permission and assisting us to conduct this study.

References

1.Balasubramani, P.P*., Walke, A., Grennan, G., Purpura, S., Perley, A., Ramanathan, D., Coleman, T., & Mishra, J. (2022). Simultaneous gut-brain electrophysiology shows cognition and satiety specific coupling. Sensors, 22(23), 9242. https://doi.org/10.3390/s22239242 *corresponding

2.Balasubramani, P. P*., Diaz-Delgado, J., Grennan, G., Alim, F., Zafar-Khan, M., Maric, V., ... & Mishra, J*. (2022). Distinct neural activations correlate with maximization of reward magnitude versus frequency. Cerebral Cortex, 2022;, bhac482, https://doi.org/10.1093/cercor/bhac482 *corresponding

Speakers

Pragathi Priyadharsini Balasubramani

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P015: Dynamic Causal Modelling in Probabilistic Programming Languages

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P015 Dynamic Causal Modelling in Probabilistic Programming Languages

Nina Baldy1*, Marmaduke Woodman1, Viktor Jirsa1, Meysam Hashemi1

1Aix Marseille Univ, INSERM, INS, Inst Neurosci Syst, Marseille, France

*Email: nina.baldy@univ-amu.fr
Introduction

Dynamic Causal Modeling (DCM) [1] is a key methodology in neuroimaging for understanding the intricate dynamics of brain activities. It imposes a statistical framework that embraces causal relationships among brain regions and their responses to experimental manipulations, such as stimulation. In this work, we perform Bayesian inference on a neurobiologically plausible model that simulates event-related potentials observed in magneto/encephalography data [2]. This translates into probabilistic inference of latent and observed states of a system described by a set of nonlinear ordinary differential equations (ODEs) and potentially correlated parameters.
Methods
Central to DCM is Bayesian model inversion, which aims to infer the posterior distribution of model parameters given the prior and observed data. Variational inference translates this into an optimization problem by approximating the posterior with a fixed-form density [3]. We consider three Gaussian approximations: the mean-field which neglects correlation between parameters, its full-rank counterpart, and the analytical Laplace. We benchmark them against state-of-the art Markov Chain Monte Carlo (MCMC): the No-U-Turn-Sampler [4]. Finally, we benchmark the efficiency of each method as implemented in several Probabilistic Programming Languages (PPLs) [5] in terms of effective sample per computational unit.

Results
Our investigation shows that model inversion in DCM extends beyond variational approximation frameworks, demonstrating the effectiveness of gradient-based MCMC. We observe close alignment between MCMC NUTS and full-rank variational in terms of posterior distributions and model comparison. Our results demonstrate significant improvements in the effective sample size per computational time unit, with PPLs showing advantages over traditional implementations. Additionally, we propose solutions to mitigate issues related to multi-modality in posterior distributions, such as initializing at the tail of the prior distribution, and weighted stacking [6] of chains for improved inference.

Discussion
Previous research on MCMC methods for Bayesian model inversion in DCM highlighted challenges with both gradient-free and gradient-based approaches [7, 8]. However, we found that the ability to combine probabilistic modeling with high-performance computational tools offers a promising solution to the challenges of high-dimensional, non-linear models in DCM. Future work should extend to whole-brain models and fMRI data, which pose additional challenges for both MCMC and variational methods.

Acknowledgements
This research has received funding from EU’s Horizon 2020 Framework Programme for Research and Innovation under the Specific Grant Agreements No. 101147319 (EBRAINS 2.0 Project), No. 101137289 (Virtual Brain Twin Project), and government grant managed by the Agence Nationale de la Recherche reference ANR-22-PESN-0012 (France 2030 program).

References
[1]https://doi.org/10.1016/S1053-8119(03)00202-7
[2]https://doi.org/10.1016/j.neuroimage.2005.10.045
[3]https://doi.org/10.1080/01621459.2017.1285773
[4]https://doi.org/10.48550/arXiv.1111.4246
[5]https://doi.org/10.1145/2593882.2593900
[6]https://doi.org/10.48550/arXiv.2006.12335
[7]https://doi.org/10.1016/j.neuroimage.2015.03.008[8]https://doi.org/10.1016/j.neuroimage.2015.07.043

Speakers

Marmaduke Woodman

Viktor Jirsa

Nina Baldy

Meysam Hashemi

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P016: Heterogeneous topologies in in silico networks are necessary to model the emergent dynamics of human-derived fully excitatory neuronal cultures

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P016 Heterogeneous topologies in in silico networks are necessary to model the emergent dynamics of human-derived fully excitatory neuronal cultures

Valerio Barabino*, 1, Francesca Callegari1, Sergio Martinoia1, Paolo Massobrio1, 2

1Department of Informatics, Bioengineering, Robotics and Systems Engineering (DIBRIS), University of Genova, Genova, Italy
2National Institute for Nuclear Physics (INFN), Genova, Italy

* Email:valerio.barabino@edu.unige.it

Introduction
Murine neuronal cultures have been the gold standard forin vitromodels, but their outcome is not always translatable to the human brain, especially in personalized medicine. Human-induced pluripotent stem cells (hiPSCs) offer a promising alternative [1]. This model requires extensive characterization, andin vitromulti-electrode arrays (MEAs) recordings alone may not capture all relevant parameters. Computational modeling can complement these experiments, offering insight into the mechanisms behind peculiar electrophysiological activities or pathological conditions [2]. This work aims to infer the underlying mechanisms behind the emergent firing profile pattern in excitatory hiPSC neuronal networks coupled to MEAs [3].
Methods
We modeled 100 Hodgkin-Huxley neurons with short-term depressing synapses. To reproduce self-sustained spontaneous activity observedin vitro, we introduced noise and external DC currents to allow for the alternation of two phases: short periods of high-frequency firing involving the whole network and long periods of asynchronous low-frequency spiking. We explored the role of external triggers, the interplay between synaptic conductances (AMPA and NMDA) and synaptic depression, and network topology in recreating the averagein vitrocumulative firing pattern. To account for the heterogeneity of biological networks, we introduced different connectivity rules, distinguishing between incoming and outgoing links.
Results
Noise emerged as the best trigger for network bursts, allowing a good balance between random spiking and bursting activity with anin vitro-like variability of inter-network burst intervals. Lower AMPA conductance than NMDA was necessary, as NMDA ensured a broader operability range forin vitro-like activity. The optimal trade-off between NMDA contribution and synaptic depression was found near a transition state, implying that small parameter changes can shift the system into different regimes. To shape cumulative firing pattern profiles, heterogeneous topologies were introduced, distinguishing afferent and efferent connectivity. The mostin vitro-like profile arose from scale-free afferent and random efferent connections.
Discussion
Consistent with previous studies [4], our findings suggest that the nature ofin vitrohiPSC network bursts is governed by a mechanism of noise amplification, controlled by a pulse of activity that is randomly nucleated and propagates throughout the network. Regarding connectivity, scale-free for afferents implies that a small subset of “privileged” neurons receives most of the inputs (hubs), thus acting as central regulators and influencing the network’s overall activity. Notably, these hubs exhibited more tonic firing, effectively acting as pacemakers that initiate network bursts, as similarly identified in [5]. However, in our case this property is structural and not an intrinsic dynamic property of single neurons.

Acknowledgements
The authors thank dr. Giulia Parodi (University of Genova) for supplying the hiPSCs recordings. This work was supported by #NEXTGENERATIONEU (NGEU) and funded by the Ministry of University and Research (MUR), National Recovery and Resilience Plan (NRRP), Project MNESYS (PE0000006)—A Multiscale integrated approach to the study of the nervous system in health and disease (DN. 1553 11.10.2022).
References
1.https://doi.org/10.1016/j.stemcr.2021.07.001
2.https://doi.org/10.1101/2024.05.23.595522
3.https://doi.org/10.1088/1741-2552/acf78b
4.https://doi.org/10.1038/nphys2686

5.https://doi.org/10.1007/s00422-010-0366-x

Speakers

Paolo Massobrio

Associate Professor, Univeristy of Genova

My research activities are in the field of the neuroengineering and computational neuroscience, including both experimental and theoretical aspects. Currently, I am coordinating a research group (1 assistant professor, 2 post-docs, and 5 PhD students) working on the interplay between... Read More →

Valerio Barabino

Francesca Callegari

Sergio Martinoia

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P017: Unraveling the neural mechanisms of behavior-related manifolds in a comprehensive model of primary motor cortex circuits

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P017 Unraveling the neural mechanisms of behavior-related manifolds in a comprehensive model of primary motor cortex circuits

Roman Baravalle1*, Valery Bragin1,5, Nikita Novikov4, Wei Xu2, Eugenio Urdapilleta3, Ian Duguid2, Salvador Dura-Bernal1,4
1 Department of Physiology and Pharmacology, SUNY Downstate Health Sciences University, Brooklyn, USA
2 Centre for Discovery Brain Sciences, University of Edinburgh, Edinburgh, UK
3 Centro Atómico Bariloche & Instituto Balseiro, Bariloche, Argentina
4 Center for Biomedical Imaging & Neuromodulation, The Nathan Kline Institute for Psychiatric Research
5 Brain Simulation Section, Charité - Universitätsmedizin Berlin, Berlin, Germany

*Corresponding Author: roman.baravalle@downstate.edu

Introduction
Accumulating evidence suggests that low-dimensional neural manifolds in the primary motor cortex (M1) play a crucial role in generating motor behavior. These latent dynamics, emerging from the collective activity of M1 neurons, are remarkably consistent across animals performing the same task. However, the specific cell types, cortical layers, and biophysical mechanisms underlying these representations remain largely unknown. Understanding these manifolds is essential for characterizing neural computations underlying behavior and has implications for developing stable and easy-to-train brain-machine interfaces (BMIs) for spinal cord injury.
Methods
We previously developed a realistic computational model of M1 circuits on NetPyNE/NEURON [1], incorporating detailed corticospinal neuron models responsible for transmitting motor commands to the spinal cord. This model was validated against in vivo spiking and local field potential data, demonstrating its ability to generate accurate predictions and provide insights into brain diseases. We further showed that M1 activity could be represented in low-dimensional manifolds, which varied according to behavioral states and experimental manipulations. These embeddings revealed clear clustering related to behavior and inactivation experiments (e.g., noradrenergic or thalamic input lesions), with high correlations between low- and high-dimensional representations.
Results
In this work, we extended the M1 model by incorporating two new interneuron types and tuning it to reproduce neural manifolds observed in vivo during a mouse joystick reaching task. Neuropixels probes recorded spiking activity in M1 and the ventrolateral thalamus, allowing us to jointly analyze neural patterns and joystick trajectories. We constructed a decoder using the CEBRA method [2] to predict movement trajectories from spiking activity and LFP and explored different model tuning strategies, including varying long-range inputs and modifying circuit connectivity via global optimization.
Discussion
Reproducing experimental behavior-related neural manifolds in large-scale cortical models enables linking neural dynamics across scales (membrane voltages, spikes, LFPs, EEG) to behavior, experimental manipulations, and disease. This approach helps refine models, characterize the relationship between latent dynamics and specific cell types, and ultimately deepen our understanding of how brain circuits generate motor behavior.
Acknowledgements
This work is supported by NIBIB U24EB028998 and NYS DOH1-C32250GG-3450000 grants

References
[1]https://doi.org/10.1016/j.celrep.2023.112574[2]https://doi.org/10.1038/s41586-023-06031-6

Speakers

Roman Baravalle

Valery Bragin

Nikita Novikov

Wei Xu

Eugenio Urdapilleta

Ian Duguid

Salvador Dura-Bernal

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P018: Orientation Bias and Abstraction in Working Memory: Evidence from Vision Models and Behaviour

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P018 Orientation Bias and Abstraction in Working Memory: Evidence from Vision Models and Behaviour

Fabio Bauer*¹, Or Yizhar¹,², Bernhard Spitzer¹,²
¹Research Group Adaptive Memory and Decision Making, Max Planck Institute for Human Development,
Berlin, Germany
²Technische Universität Dresden, Dresden, Germany*bauer@mpib-berlin.mpg.de
Introduction

Working memory (WM) for visual orientations shows behavioral bias, where remembered orientations are repelled from the cardinal axes. These canonical biases are well-documented for grating stimuli within 180° space[1-4]. WM maintenance of orientation information has been shown to involvelower-level visual processing [5-9]. However, in recent work we showed that orientation biases are also found with real-world objects in 360° space, which points to a high level of abstraction[10]. Can such abstraction and bias be explained by visual processing alone? Here, we examine if orientation biases for real-world objects emerge in computer vision models of the ventral visual stream [11,12] and compare them with behavioral reports in a WM task.
Methods
We compared activations from a range of neural network models: brain-inspired CNNs[13], established feedforward CNNs[14-16], and vision-transformers[17, 18]. Each model was shown 144 natural objects with a different principle axis (not rotationally symmetric), rotated in 16 orientations spanning 360°. We used representational similarity analysis to compare the models’ layer activations to idealized representations of bias in 180° and 360° orientation space. Results were compared with human behavioral reports from orientation WM tasks with the same kind of stimuli.
Results
Neural networks showed orientation biases in 180° space, which became stronger in deeper layers that have been suggested to model higher visual areas. In contrast, when analyzing the full 360° orientation space with natural objects, these same models showed no orientation bias at any layer. This failure across architectures reveals a fundamental limitation: while models can process orientation relationships in simple symmetric stimuli, they fail to recognize that differently shaped objects (like horizontal tables versus vertical towers) can share the same orientation. Our parallel human behavioral experiment showed that, unlike these models, people show orientation biases in working memory across the full 360° spectrum with similar natural objects.
Discussion
We found no evidence for a biased representation in 360° space in any layers of the vision models we tested. In contrast, human behavioral reports and eye-gaze patterns from WM experiments did show a clear 360° bias. This indicates that bias in our task emerges at the level of an abstracted stimulus feature (the object’s orientation relative to its real-life upright position), rather than low-level visual features. Our findings also suggest that with such real-world objects requiring abstraction, 360° orientation information is not represented in these most current models of visual processing. Future work should focus on validating these exploratory findings experimentally.

Acknowledgements
We acknowledge the Max Planck Institute for Human Development for providing computing resources and facilities. We also thank the International Max Planck Research School on Computational Methods in Psychiatry and Ageing Research (IMPRS COMP2PSYCH) for funding support. Additionally, we appreciate helpful discussions and comments from Felix Broehl and Ines Pont Sanchis.
References

1doi.org/10.1037/h0033117
2doi.org/10.1038/nn.2831
3doi.org/10.1167/10.10.6
4doi.org/10.1016/j.visres.2009.12.005
5doi.org/10.1038/nature07832
6doi.org/10.7554/eLife.94191.3
7doi.org/10.1371/journal.pbio.3001711
8doi.org/10.1080/13506285.2021.1915902
9doi.org/10.1101/2023.05.18.541327
10doi.org/10.1038/s41562-023-01737-z
11doi.org/10.1073/pnas.1403112111
12doi.org/10.1038/s41467-024-53147-y
13doi.org/10.1101/408385
14doi.org/10.1145/3065386
15doi.org/10.1109/cvpr.2017.634
16doi.org/10.48550/ARXIV.1409.155617doi.org/0.48550/ARXIV.2112.127501

18doi.org/10.48550/arXiv.2010.11929

Speakers

Fabio Bauer

Or Yizhar

Bernhard Spitzer

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P019: Distinguishing spatiotemporal scales within a connectome reveals integration and segregation efficiency in global patterns of neuronal activity

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P019 Distinguishing spatiotemporal scales within a connectome reveals integration and segregation efficiency in global patterns of neuronal activity

Diego Becerra*1,2, Ignacio Ampuero1,2, Pedro Mediano3, Christopher Connor4, Andrea Calixto2, & Patricio Orio1,2

1 Valparaíso Neural Dynamics Laboratory, Faculty of Sciences, Universidad de Valparaíso, Chile
2 Centro Interdisciplinario de Neurociencia de Valparaíso, Universidad de Valparaíso, Chile
3 Department of Computing, Imperial College London, United Kingdom
4 Brigham & Women’s Hospital, Harvard Medical School, Boston, USA.

* Email: becerra.q.diego@gmail.com
Introduction

Neurons in the brain communicate in different ways, and thus connectomes can be conceived as overlapping dissimilar networks depending on the type of signal being transmitted. One crucial difference between modes of signalling is given by the connectivity timescales, yielding four layers of paths (ordered from fastest to slowest): gap junctions, amino acid, monoaminergic, and peptidergic transmitters.Caenorhabditis elegans, a 302-neuron nematode, is an excellent model for exploring both topological and functional properties of the interaction between layers of networks, because its full connectome is known, alongside a lot of its electrophysiology.

Methods
A full structural connectome ofC. eleganswas built from the latest empirical works available. Functional connectomes were built from twoC. elegans‘whole-brain’ calcium imaging datasets of global states: npr-1mutants undergoing quiescence and wakefulness [2]; and QW1217 mutants undergoing 4% and 8% of isoflurane anesthesia [1].
We analyzed integration and segregation applying network topology measures to the datasets and to the connectome layers. Also, we developed a partial network decomposition [PND] algorithm which analyzes the shortest path between nodes of a pair of overlapping networks. We then compared the network properties of theC. elegansconnectomes with lattiziced and randomized surrogates.
Results
While the peptidergic connectome is dense, the others are sparse. Applying PND, we determined if a path between nodes is redundant, uniquely contributed, or synergistic for a pair of spatiotemporal scales. Unique paths are predominant in all pairs of scales, the highest redundancy is found between electrical and amino acid transmission, and the highest synergy is between electrical and monoaminergic.

Empirical pairs of connectomes are more synergistic than their latticized or randomized surrogates, suggesting that the empirical network yields an improvement in efficiency. Comparing segregation and integration between structural (SC) and functional (FC) connectivity, FC of asleep and anesthetized worms is closer to SC than the FC of awake worms.
Discussion
We were able to characterize complementary (synergistic), redundant, and unique paths between nodes of the connectomes. Yet, the recent integration of gene-expression datasets and ligand-receptor interaction shows a pervasive extra-synaptic transmission network. To discover the effect of differences in connectivity density between peptidergic and the other scales of neurotransmission require thus including the temporal dimension: both by using empirical ‘whole-brain’ datasets portraying different global states (wakefulness, sleep, anesthesia) and a mathematical model using the full topology of the network, which is in process of being implemented.

Figure 1. Fig. 1. (A) Center: Shortest paths of the empirical connectome layers favor synergy as distance between nodes increase, when compared to latticized (left) and randomized (right) versions. Binarizing top Pearson correlations of awake vs. asleep (B); and anesthetized vs. awake (C) timeseries show that structural segregation (left cols.) and integration (right cols.) values are closer to asleep ones.
Acknowledgements
Fondo Nacional de Desarrollo Científico y Tecnológico (FONDECYTPatricio Orio grant number 1241469; ANID-Basal: Patricio Orio grant number AFB240002; ANID-Doctoral Fellowship: Diego Becerra 21210914.
References
1. Chang, A. S., Wirak, G. S., Li, D., Gabel, C. V., & Connor, C. W. (2023).Measures of Information Content during Anesthesia and Emergence in the Caenorhabditis elegans Nervous System.Anesthesiology, 139(1), 49–62.https://doi.org/10.1097/ALN.0000000000004579
2. Nichols, A. L. A., Eichler, T., Latham, R., & Zimmer, M. (2017).A global brain state underlies C. Elegans sleep behavior.Science, 356(6344), 1247–1256.https://doi.org/10.1126/science.aam6851

Speakers

Ignacio Ampuero

Patricio Orio

Diego Becerra

Pedro Mediano

Christopher Connor

Andrea Calixto

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P020: A Multi-Compartment Computational Approach to Cerebellar Circuit Dysfunction in Autism

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P020 A Multi-Compartment Computational Approach to Cerebellar Circuit Dysfunction in Autism

Danilo Benozzo*1, Martina F. Rizza1, Danila Di Domenico1, Giorgia Pellavio1, Filippo Marchetti1, Egidio D’Angelo1,2, Claudia Casellato1

¹ Department of Brain and Behavioural Sciences, University of Pavia, Pavia, Italy
² Digital Neuroscience Center, IRCCS Mondino Foundation, Pavia, Italy

*Email: danilo.benozzo@unipv.it
Introduction

Modeling brain dynamics requires addressing processes that span different temporal and spatial scales [1]. This is crucial, when studying phenomena at the network level that are consequences of pharmacological or pathological alterations occurring at the single-cell level, such as changes in ionic or synaptic currents. Our aim is to study how single-cell dynamics affect circuit dynamics in a mouse model of autism (IB2 knock-out, KO), within the context of the cerebellar cortical microcircuit. Cerebellar implications in autism spectrum disorders (ASD) have been well-documented, showing a dependent association between cerebellar damages and an increased risk for ASD [2].
Methods
We re-parameterized a wild-type (WT) granule cell (GrC) multi-compartment model [3] to match the empirical properties of IB2-KO GrCs [4]. At the network level, we employed a bottom-up approach, by placing all the cell types that characterize the cerebellar cortex, preserving their physiological morphology, density and connection affinity. On the simulation side, the activity of each cell type was reproduced through a multi-compartment model interfaced with the NEURON simulator [5]. The entire process was managed by the Brain Scaffold Builder framework [6,7].
Results
In the WT GrC model we increased Na and K maximum conductances (gmax) to match the higher in/outward currents in IB2 GrCs. Tonic glutamate levels [glu] in mossy-fiber-GrC synapses and NMDA gmax were adjusted to replicate experimental I-f and NMDA currents, predicting [glu] at 11.2 µM and a 4x NMDA gmax increase. The IB2 GrC model was integrated into the canonical cerebellar circuit, assuming no other cell changes (empirical IB2 Purkinje cell (PC) spks/s = 51.8, std 11.7, no sign. vs. WT). Network comparisons revealed greater stimulus spread through the Gr-layer, Fig.1B. Fig.1C shows peri-stimulus histograms for both circuits under different input, predicting an overall firing increase (rates from 9.5x in GrCs to 1.6x in PCs).
Discussion
This bottom-up modelling framework enabled us to construct a representative microcircuit of the mouse cerebellar cortex, featuring a granular layer that replicates the alterations empirically observed in the IB2-KO model. This multiscale approach allows us to predict how the circuit dynamics respond to single-cell model modifications. In the granular layer, our results reflect the spatially expanded, higher E/I balance around IB2-KO GrCs observed in [4]. To further validate whole-circuit activity, we are currently comparing our predictions with in vitro MEA recordings from both WT and IB2-KO mice, in spontaneous regime and under mossy-fiber impulse stimulations.

Figure 1. A: Effect of NMDA gmax (kgmax NMDA) and ambient glutamate ([glu]) on NMDA currents in IB2-KO GrCs. B: How a 20 Hz stim propagates through the granular layer, applied to mossy fibers (mfs) within the circular target, r=40 µm. C: Firing rates of each cell type under three conditions: no stimulus, an 8 Hz Poisson basal input to all mfs, and basal input plus high-frequency stim targeted to 15 mfs.
Acknowledgements
Work supported by NEXTGENERATIONEU (NGEU) and funded by the Ministry of University and Research (MUR), National Recovery and Resilience Plan (NRRP), project MNESYS (PE0000006) – A Multiscale integrated approach to the study of the nervous system in health and disease (DN. 1553 11.10.2022)
References
[1]https://doi.org/10.1162/netn_a_00343
[2]https://doi.org/10.1016/j.ijdevneu.2004.09.006
[3]https://doi.org/10.3389/fncel.2017.00071
[4]https://doi.org/10.1523/jneurosci.1985-18.2019
[5]https://doi.org/10.1017/cbo9780511541612
[6]https://doi.org/10.1038/s42003-022-04213-y[7]https://ebrains.eu/service/brain-scaffold-builder/

Speakers

Claudia Casellato

Danilo Benozzo

Martina Rizza

Danila Di Domenico

Giorgia Pellavio

Filippo Marchetti

Egidio D'Angelo

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P021: Studying the effects of TMS on the cerebellum with a realistic model of the cerebellar cortex

P026: Prospective and retrospective coding in cortical neurons

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P026 Prospective and retrospective coding in cortical neurons

Simon Brandt1, Paul Haider*,1, Mihai A. Petrovici1, Walter Senn1, Katharina A. Wilmesa,1, Federico Beniteza,1

1Department of Physiology, University of Bern, Switzerland
ashared senior authorship

*Email: paul.haider@unibe.ch

Introduction
Brains can process sensory information from different modalities at astonishing speed; this is surprising as already the integration of inputs through the membrane causes a delayed response Fig. 1d. Experiments reveal a possible explanation for the fast processing, showing that neurons can advance their output firing rate with respect to their input current, a concept which we refer to as prospective coding [1]. The combination of retrospective (a delayed response) and prospective coding enables neurons to perform temporal processing. While retrospective coding emerges from the inherent delays of neurons, the underlying mechanisms of prospective coding, however, are not completely understood.

Methods
In this work, we elucidate cellular mechanisms which can explain prospective firing in cortical neurons. We use simulation-based inference to investigate the parameters of the Hodgkin-Huxley model [2] with respect to its ability to fire prospective and retrospective. Based on this analysis, we derive a reduced model that allows us manipulate the temporal response of the Hodgkin-Huxley neurons. Further on, we derive rate based models of neurons which include adaption processes on arbitrary time scales to investigate advances on longer time scales.

Results
We show that the spike generation mechanism can be the source for the prospective (advanced) or retrospective (delayed) response as shown for prospective firing (Fig. 1a) in cortical-like neurons [3,4] (Fig. 1b, green) and retrospective firing (Fig. 1c) in hippocampal-like neurons [5,6] (Fig. 1b, orange). Further, we analyse the Hodgkin-Huxley dynamics to derive a reduced model to manipulate the timing of the neuron’s output by tuning three parameters (Fig. 1d-h). We further show that slow adaptation processes, such as spike-frequency adaptation or deactivating dendritic currents, can generate prospective firing for inputs that undergo slow temporal modulations. In general, we show that adaptation processes at different time scales can cause advanced neuronal responses to time-varying inputs that are modulated on the corresponding time scales.

Discussion
The results of this work contribute to the understanding of how fast processing (prospective coding) and short-term memory (retrospective coding) can be achieved in the brain on the level of single neurons and might guide further experiments. Prospectivity and retrospectivity may be important for several cognitive functions. The interplay of the two provides a powerful framework for temporal processing by shifting signals in time. The insights are highly beneficial for biologically plausible learning algorithms used for temporal processing and their implementation on neuromorphic hardware [7-9].
Figure 1. (a) Hodgkin-Huxley neurons can be prospective for cortical neurons (b, green) and retrospective (c) for parameters fitted to hippocampal neurons (b, orange). (d) Because a neuron integrates input through its membrane, a response of the neuron is expected to be delayed by the membrane. If the output of a neuron can be advanced with respect to its input, a prospective mechanism needs to exist. With
Acknowledgements
We would like to express particular gratitude for the ongoing support from the Manfred Stärk Foundation. Our work has greatly benefited from access to the Fenix Infrastructure resources, which are partially funded through the ICEI project under the grant agreement No. 800858. This includes access to Piz Daint at the Swiss National Supercomputing Centre, Switzerland.
References

1.https://doi.org/10.1093/cercor/bhm235
2.https://doi.org/10.1113/jphysiol.1952.sp004764
3. https://doi.org/10.1017/CBO9781107447615
4. https://doi.org/10.1016/0896-6273(95)90020-9
5.https://doi.org/10.1007/s10827-007-0038-6
6.https://doi.org/10.1017/CBO9780511895401
7.https://doi.org/10.7554/elife.89674
8.https://doi.org/10.48550/arXiv.2110.14549
9. https://doi.org/10.48550/arXiv.2403.16933

Speakers

Paul Haider

Simon Brandt

Mihai A. Petrovici

Walter Senn

Katharina Anna Wilmes

Federico Benitez

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P027: A functional network model for body column neural connectivity in Hydra

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P027 A functional network model for body column neural connectivity in Hydra

Wilhelm Braun*1,2, Sebastian Jenderny4, Christoph Giez5,6, Dijana Pavleska5, Alexander Klimovich5,

Thomas C.G. Bosch5, Karlheinz Ochs4, Philipp Hövel7, Claus C. Hilgetag1,3

1Institute of Computational Neuroscience, Center for Experimental Medicine, University Medical Center Hamburg-Eppendorf, Martinistraße 52, 20246, Hamburg, Germany

2 Faculty of Engineering, Department of Electrical and Information Engineering, Kiel University, Kaiserstraße 2, 24143, Kiel, Germany

3Department of Health Sciences, Boston University, 635 Commonwealth Avenue, Boston, MA, 02215, USA

4Chair of Digital Communication Systems, Ruhr-Universität Bochum, Universitätsstraße 150, 44801, Bochum, North Rhine-Westphalia, Germany

5Zoological Institute, University of Kiel, Christian-Albrechts-Platz 4, 24118 Kiel, Germany

6The Francis Crick Institute, London NW1 1BF, UK

7Theoretical Physics and Center for Biophysics, Saarland University, Campus E2 6, Saarbrücken, 66123, Germany

*Email: wilhelm_braun@icloud.com

Introduction

Hydrais a non-senescent animal with a relatively small number of cell types and overall low structural complexity, but a surprisingly rich behavioral repertoire. The main drivers ofHydra’s behavior are neurons that are arranged in two nerve nets comprising several distinct neuronal populations. Among these populations is the ectodermal nerve net N3 which is located throughout the animal. It has been shown that N3 is necessary and sufficient for the complex behavior of somersaulting [1] and is also involved inHydrafeeding behavior [2, 3]. Despite being a behavioral jack-of-all-trades, there is insufficient knowledge on the coupling structure of neurons in N3, its connectome, and its role in activity propagation and function.

Methods
We construct a model connectome for the part of N3 located on the body column by using pairwise distance- and connection angle-dependent connectivity rules. Using experimental data on the placement of neuronal somata and the spatial dimensions of the body column, we design a generative network model combining non-random placement of neuronal somata and the preferred orientation of primary neurites. Additionally, we study activity propagation in N3 using the simple excitable Susceptible-Excited-Refractory (SER) model and a more complex neuromorphic Morris-Lecar model.

Results
We show [4] that the generative network model yields good agreement with experimental data. We then show that the simple excitable dynamical SER model generates directed, short-lived, fast propagating patterns of activity. In addition, by slightly changing the parameters of the dynamical model, the same structural network can also generate persistent activity. Finally, we use a neuromorphic circuit based on the Morris-Lecar model to show that the same structural connectome can, in addition to through-conductance with biologically plausible time scales, also host a dynamical pattern related to the complex behavioral pattern of somersaulting.

Discussion
Our work provides a systematic construction of the structure of a subnetwork ofHydra’snervous system. By assuming that the ectodermal body column network inHydrais essentially two-dimensional, we designed a generative network model that is in agreement with measured
structural quantities and supports two different activity modes, each presumably controlling
different types of behavior inHydra. We speculate that such different dynamical regimes act as dynamical substrates for the different functional roles of N3, allowingHydrato exhibit behavioral complexity with a relatively simple nervous system that does not possess modules or hubs.

Acknowledgements
WB would like to thank Kayson Fakhar, Fatemeh Hadaeghi and Mariia Popova for helpful discussions. Funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) – Project-ID 434434223 – SFB 1461.

References
[1]10.1016/j.cub.2023.03.047
[2]10.1016/j.cub.2023.10.038
[3]10.1016/j.celrep.2024.114210
[4]10.1101/2024.06.25.600563

Speakers

Wilhelm Braun

Junior Research Group leader, CAU Kiel, Department of Electrical and Information Engineering

Claus Hilgetag

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P028: The Topological Significance of Functional Connectivity State Transitions

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P028 The Topological Significance of Functional Connectivity State Transitions

Spencer Brown¹, Celine Zalamea², DanielSelski, PhD3

¹ College of Osteopathic Medicine, Pacific Northwest University of Health Sciences, Yakima, United States of America
² College of Osteopathic Medicine, Pacific Northwest University of Health Sciences, Yakima, United States of America
³College of Osteopathic Medicine, Pacific Northwest University of Health Sciences, Yakima, United States of America

Email:smbrown@pnwu.edu

Introduction

Pathology in dynamic functional connectivity is well documented but lacks explanatory rationale.Dynamicfunctional connectivityparallels the existence of discrete state transitions,the study ofwhich may provide further elucidation. In this study, weutilizedTopological Data Analysis (TDA) toobservethe shape of whole brain networks and the transitions between them. Further, we aim to understand thestructure-function relationship of brain states with neuronal physiology.
Methods
fMRI scans were obtainedand preprocessedfrom the Human Connectome Project motor task dataset.States were definedrelativeto an anticipatory visual cue. We initially used the Euclidian distance (L2) toclusterstateshierarchically.We converted brain states to Vietoris-Rips complexestoidentifytopology.Distances between these complexes were then measured using the Wasserstein distanceandaggregatedusinghierarchicalclustering.
Results
Wedemonstratethateach L2 state label mustonly havea single topology, but the same topology may exist in multiple states.Under this assumption, many combinations of L2 and topology wereobservedto be invalid.Thisisreconciled by anintrinsichierarchy of brain states.
Discussion
Weobservethatbrainstates maybedrastically different networks but share the same topology. For instance,resting versus task states mayexhibitthesame topology.In contrast, similar states may also differ in topology.We findthat topology may have a unique role in neuronal physiology andprovidesa potential framework for further studies that explorebrain dynamics.

Acknowledgements
N/A.
References
N/A.

Speakers

Spencer Brown

Celine Zalamea

Daniel Selski

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P029: Spiking Neural Networks for Controlling a Biomechanically Realistic Arm Model

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P029 Spiking Neural Networks for Controlling a Biomechanically Realistic Arm Model

Philip Bröhl*1,2, Junji Ito2, Ira Assent1,3, Sonja Grün2,4,5

1Institute for Advanced Simulation (IAS-8), Research Center Juelich, 52425 Jülich, Germany
2 Institute for Advanced Simulation (IAS-6), Research Center Juelich, 52425 Jülich, Germany
3Department of Computer Science, Aarhus University, 8200 Aarhus N, Denmark
4JARA Institute Brain Structure Function Relationship, Research Center Juelich, 52425 Jülich, Germany
5 Theoretical Systems Neurobiology, RWTH Aachen Univ., Aachen, Germany

*Email: p.broehl@fz-juelich.de
Introduction

A typical feature of neurons in the motor cortex of mammalian brains is that they are tuned to a particular direction of movements, i.e. they exhibit most spikes when a body part is moved in a particular direction, called preferred direction (PD). It has been reported that the distribution of preferred directions among motor cortex neurons depends on the constraints in the movements: when the arm may move freely in 3D, it is uniform [1,2], but when it is constrained to a 2D movement, it is bimodal [3,4]. In this work, we aim at revealing the neuronal mechanism underlying the emergence of a bimodal PD distribution, by studying an artificial network of spiking neurons trained to control a biomechanically realistic arm model.
Methods

Our model is implemented in Tensorflow [5] and consists of 300 recurrent leaky integrate-and-fire neurons with 6 linear readout neurons that control the 6 muscles in a biomechanical arm model [4]. We train it to output muscle activation signals to perform a 2D reaching task. We study its output space by applying a Principal Component Analysis on the outputs and relate the directions in this space to the directions in the joint angle arm acceleration space via Canonical Correlation Analysis. We also study the effect of each recurrent neuron on the output dynamics by interpreting its outgoing connection weights as a direction in the space of the recurrent dynamics and projecting this direction onto the output space via the readout weights.
Results

The model neurons show directional tuning with bimodally distributed PDs. The output dynamics of the model are well captured by the first two principal components (PCs). The first PC aligns to the two opposite directions in the joint angle acceleration space which agree with the hand movement directions corresponding to the peaks of the bimodal PD distribution. The effects of neurons on the output dynamics concentrate around these directions. Connections between neurons with similar output effects tend to be strongly excitatory. Taken together, the core architecture of the recurrent network is characterized by two clusters of neurons with strong excitatory connections in each cluster. Connections between the clusters are mostly inhibitory.
Discussion

The analysis shows that two mutually inhibiting clusters of excitatory connections underlie the control of the biomechanical arm model in a 2D reaching task by a recurrent network of spiking neurons. Since each of the two clusters is composed of neurons with similar output effect directions, which we have shown to be related to hand movement directions, the existence of the two clusters naturally explains the bimodality of the hand movement PD directions. This leads to the question whether similar structures are employed in the mammalian brain to control movements, which would be subject to future research.

Acknowledgements

This work was partially performed as part of the Helmholtz School for Data Science in Life, Earth and Energy (HDS-LEE) and received funding from the Helmholtz Association of German Research Centres. This research was partially funded by the NRW-network 'iBehave', grant number: NW21-049.

References

1. https://doi.org/10.1523/JNEUROSCI.10-07-02039.1990
2. https://doi.org/10.1523/JNEUROSCI.08-08-02913.1988
3.https://doi.org/10.1016/j.neuron.2012.10.041
4. https://doi.org/10.7554/eLife.88591.3
5. https://doi.org/10.5281/zenodo.4724125

Lorenzo Butti

Nathaniel Powell

Bettina Hein

Jonas Elpelt

Haleigh Mulholland

Gordon Smith

Matthias Kaschube

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P033: Modeling Corticospinal Input Effects on V1-FoxP2 Interneurons in a Go Task Using NetPyNE

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P033 Modeling Corticospinal Input Effects on V1-FoxP2 Interneurons in a Go Task Using NetPyNE

Andres F. Cadena Parra1*, Michelle Sanchez Rivera3, Constantinos Eleftheriou3, Roman Baravalle2, Ian Duguid3, Salvador Dura-Bernal2,4
1Department of Biomedical Engineering. Universidad de los Andes, Bogota, Colombia
2Department of Physiology and Pharmacology, SUNY Downstate Health Sciences University, Brooklyn, USA
3Centre for Discovery Brain Sciences & Simons Initiative for the Developing Brain, University of Edinburgh, Edinburgh, UK.
4Center for Biomedical Imaging & Neuromodulation, The Nathan Kline Institute for Psychiatric Research.

*Email: af.cadena@uniandes.edu.co
Introduction

The execution of movement involves a complex interplay of neural structures whose activity results in precise motor output. The motor cortex generates commands for voluntary movement, conveyed via corticospinal neurons (CSNs) to the spinal cord, where interneurons (INs) integrate sensory feedback and motor commands to fine-tune motor neuron activity [1]. Among these V1 INs, and particularly those expressing FoxP2, are crucial for inhibitory feedback in motor control [2]. Building on recent findings that a subset of CSNs exhibit decreased firing rates during movement, this study aims to investigate how these CSN inputs influence V1-FoxP2 interneurons, shedding light on spinal integration mechanisms for coordinated motor output.
Methods
An in silico model was developed to study the effects of convergent increased/decreased corticospinal input on V1-FoxP2 INs. A single-cell model was implemented in NetPyNE/NEURON, incorporating Na⁺, K⁺, Ca²⁺ channels and AMPA dynamics. The cell’s morphology had a soma and four dendritic sections. Calibration used in vitro current-clamp data and optimization. Spike trains from a “Go task” from two different CSN subpopulations were connected to the V1-FoxP2 model, with simulated background activity consistent with in vivo observations [3]. Three conditions were tested: (1) increasing/decreasing input, (2) increasing/sustained input, and (3) increasing input only. Electrophysiological properties, like input resistance, were recorded over time.
Results
The model simulated in vivo V1-FoxP2 dynamics, where corticospinal input initially drives an increase in firing rate that peaks at movement onset, followed by a return to baseline. The in vivo condition optimizes the input-output relationship, exhibiting a high signal-to-noise ratio (SNR) post-movement and enabling a quicker return to baseline excitability. Additionally, background activity enhances the return to baseline of the V1-FoxP2 firing rate and input resistance. Notably, input resistance decreased progressively across time windows before, during, and after movement, making the neuron less susceptible to noise. The model further revealed that movement requires a specific ratio of increased and decreased CSN inputs.
Discussion
Our findings provide key insights into the neuronal computations that govern the integration of cortical inputs in the spinal cord. The model showed that in vivo-like corticospinal input enhances V1-FoxP2 activity time-locking to behavior without significantly reducing the SNR. This indicates a trade-off between temporal precision and firing rate strength to optimize motor control. Additionally, decreasing CSN input facilitates impedance recovery after movement, whereas in the sustained scenario, impedance fails to return to baseline. These results may inform future studies on the functional architecture of spinal circuits involved in motor control and rehabilitation, particularly in disorders affecting motor coordination.

Acknowledgements
This work was supported by the NIBIB U24EB028998 and NYS DOH01-C32250GG-3450000 grants. AFCP was supported by Universidad de los Andes through a Teaching Assistantship.
References
[1] Deska-Gauthier, D., & Zhang, Y. (2019). Functional diversity of spinal interneurons and locomotor control. Curr. Opin. Physiol., 8, 99–108. https://doi.org/10.1016/j.cophys.2019.01.005
[2] Bikoff, J. B., Gabitto, M. I., Rivard, A. F., et al. (2016). Spinal inhibitory interneuron diversity delineates motor microcircuits. Cell, 165, 207–219. https://doi.org/10.1016/j.cell.2016.01.027
[3] Schiemann, J., Puggioni, P., Dacre, J., et al. (2015). Behavioral state-dependent modulation of motor cortex output. Cell Rep., 11, 1319–1330.https://doi.org/10.1016/j.celrep.2015.04.042

Speakers

Roman Baravalle

Ian Duguid

Salvador Dura-Bernal

Andres Felipe Cadena Parra

Michelle Sanchez Rivera

Constantinos Eleftheriou

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P034: Role of Synaptic Plasticity in the Emergence of Temporal Complexity in a Izhikevich Spiking Neural Network

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P034 Role of Synaptic Plasticity in the Emergence of Temporal Complexity in a Izhikevich Spiking Neural Network

Marco Cafiso*1,2, Paolo Paradisi2,3

1Department of Physics 'E. Fermi', University of Pisa, Largo Bruno Pontecorvo 3, I-56127, Pisa, Italy
2Institute of Information Science and Technologies ‘A. Faedo’, ISTI-CNR, Via G. Moruzzi 1, I-56124, Pisa, Italy
3BCAM-Basque Center for Applied Mathematics, Alameda de Mazarredo 14, E-48009, Bilbao, BASQUE COUNTRY, Spain

*Email: marco.cafiso@phd.unipi.it
Introduction

Neural avalanches exemplify intermittent behavior in brain dynamics through large-scale regional interactions and are crucial elements of brain dynamical behaviors. Originally introduced in the Self-Organized Criticality framework, these intermittent complex behaviors can also be examined through Temporal Complexity (TC) theory. Computational neural network models have become central in the neuroscience field. Izhikevich’s neuron model [1] provides a powerful yet simple framework for simulating networks with over 20 brain-like dynamic patterns, enabling studies of normal and pathological conditions. Our work analyzes the temporal complexity of neural avalanches and coincidence events in an Izhikevich Spiking Neural Network, comparing systems with and without Spike-Time Dependent Plasticity (STDP) [2] processes.
Methods
A network of 1,000 Izhikevich neurons with an excitatory-to-inhibitory ratio of 4:1 was developed, designing inhibitory synaptic connections to exert a stronger influence than their excitatory counterparts, reflecting physiological neural circuit dynamics. We subjected the network to six distinct input signals, including two containing complex events. We then measured and compared the temporal complexity of network responses both with and without STDP plasticity mechanisms activated. Our temporal complexity assessment methodology leverages neural avalanche and coincidence events to estimate multiple scaling indices [3]. These metrics provide quantitative measures of the system’s complexity.
Results
The analysis of scaling indices related to temporal complexity reveals variations in complexity within neural avalanches and coincidences in simulations that incorporate the STDP plasticity rule, compared to those where it is absent. Furthermore, the extent of the change in temporal complexity depends on the simulation’s input signal. Specifically, strong and continuous signals lead to a substantial change in temporal complexity when the STDP rule is present, whereas intermittent signals exhibit smaller variations in complexity due to STDP.
Discussion
These preliminary results on the complexity behaviors of a spiking neural network with or without the STDP plasticity rule highlight how topological changes in the network configuration, due to time-dependent plasticity rules, lead to changes in temporal complexity behaviors. These results suggest that neural plasticity, defined as changes in the network’s spatial configuration, can influence the temporal complexity levels of a neuronal network, providing insights into the dynamic interplay between structural adaptation and the emergence of temporal complex behaviors in spiking neural networks.

Acknowledgements
This work was supported by the Next-Generation-EU programme under the funding schemes PNRR-PE-AI scheme (M4C2, investment 1.3, line on AI) FAIR “Future Artificial Intelligence Research”, grant id PE00000013, Spoke-8: Pervasive AI.
References
[1] Eugene M. Izhikevich. “Simple model of spiking neurons”. In: IEEE Transactions on Neural Networks 14.6 (2003), pp. 1569–1572.
[2] Natalia Caporale and Yang Dan. “Spike Timing–Dependent Plasticity: A Hebbian Learning Rule”. In: Annual Review of Neuroscience 31.Volume 31, 2008 (2008), pp. 25–46.
[3] P. Paradisi and P. Allegrini. “Intermittency-Driven Complexity in Signal Processing”. In: Complexity and Nonlinearity in Cardiovascular Signals. Ed. by Riccardo Barbieri, Enzo Pasquale Scilingo, and Gaetano Valenza. Cham: Springer, 2017, pp. 161–195.

Speakers

Marco Cafiso

Paolo Paradisi

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P035: The geometry of primary visual cortex representations is dynamically adapted to task performance

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P035 The geometry of primary visual cortex representations is dynamically adapted to task performance

Leyla Roksan Caglar*1,Julien Corbo*2, O.Batuhan Erkat2,3, Pierre-Olivier Polack2

1Windreich Department of AI & Human Health, Icahn School of Medicine at Mount Sinai, New York, NY, USA
2Center for Molecular and Behavioral Neuroscience, Rutgers University—Newark, Newark, NJ, USA
3Graduate Program in Neuroscience, Rutgers University—Newark, Newark, NJ, USA

*Contributed equally; Email: l.r.caglar@gmail.com; julien.corbo@gmail.com
Introduction

Perceptual learning optimizes perception by reshaping sensory representations to enhance discrimination and generalization. Although these mechanisms’ implementation remains elusive, recent advances suggest that the neural geometry of the representations is key, by preparing population activity to be read out at the next processing stage. Our previous work has shown that learning a visual discrimination task reshapes the population feature representations in the primary visual cortex (V1) via suppressive mechanisms, effectively discretizing the representational space, and favoring categorization and generalization [1]. However, it is unclear how these changes impact the discriminability of the representations when being read out and transformed into a decision variable.
Methods
Recent findings under the Manifold Capacity Theory [2]suggest that learning enhances classification capacity by altering the geometric properties of population activity, increasing the linear separability of stimulus representations. To test this, we examined the relationship between V1 feature representation, neural manifold geometry, and behavioral discrimination, hypothesizing that the previously observed discretization would enhance classification capacity and alter manifold geometry as early as V1. Using calcium imaging, we compared V1 activity between trained and naïve mice performing an orientation discrimination Go/NoGo task at varying difficulty levels.
Results
Investigating response dimensionality, we found it increased as the Go/NoGo stimuli became more similar in both trained and naïve mice. As predicted, dimensionality was lower in trained animals, suggesting the task's biological implementation relies on reducing representational dimensionality. However, dimensionality alone did not fully explain performance variability. Instead, we found that the linear separability of representations in their embedding space was a stronger predictor of individual behavioral performance. This separability of manifolds was further evidenced by measuring the neural manifold’s capacity and their geometric properties (manifold dimension and manifold radius), which all show a decrease with successful behavioral performance in the trained mice, but show no change in the naive mice.
Discussion
Taken together, our results show a clear relationship between behavioral task performance, representational dimensionality, and manifold separability in the early visual cortex of mice. Across all computational measures, we demonstrated an inverse relationship between dimensionality and successful perceptual discrimination assisted by representational separability.These results confirm that learning alters the geometric properties of early sensory representations as early as in V1, optimizing them for linear readout and improving perceptual decision-making.

Acknowledgements
The authors are grateful to the members of the Polack lab for the helpful conversations. This work was funded by The Whitehall Foundation (grant 2015-08-69). The Charles and Johanna Busch Biomedical Grant Program The National Institutes of Health National Eye Institute: Grant #R01 EY030860 Brain initiative: Grant #R01 NS120289) Fyssen Foundation postdoctoral fellowship.
References
[1] Corbo, J., Erkat, O. B., McClure, J., Khdour, H., & Polack, P.-O. (2025). Discretized representations in V1 predict suboptimal orientation discrimination.Nature Communications,16(1), 41. https://doi.org/10.1038/s41467-024-55409-1
[2]Chung, S., Lee, D. D., & Sompolinsky, H. (2016). Linear readout of object manifolds.Physical Review E,93(6), 060301. https://doi.org/10.1103/PhysRevE.93.060301

Speakers

Leyla Roksan Caglar

Julien Corbo

O. Batuhan Erkat

Pierre-Olivier Polack

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P039 In Silico Safety and Performance Assessment of Vagus Nerve Stimulation with Metamodeling-Based Uncertainty/Variability Quantification

Antonino M. Cassara'*1, Javier Garcia Ordonez1, Werner Van Geit1, Esra Neufeld1

1Foundation for Research on Information Technologies in Society (IT’IS), Zurich, Switzerland

*Email: cassara@itis.swiss

Introduction

Safety and efficacy assessments of medical devices are key to regulatory submissions. We established anin silicopipeline for neural interface assessment and demonstrated it for a vagus nerve stimulation (VNS) cuff electrode. It combines histology-based electromagnetic, electrophysiology, and thermal simulations, as well as tissue damage predictors, with high throughput screening of data from the NIH SPARC program [1], and systematic uncertainty quantification to assess safety and shed light on primary concerns, dominant factors, variability, and model limitations. This study serves to guide the development and application of regulatory-gradein silicomethodologies for safer, more effective medical technologies.

Methods
Evaluated quantities-of-interest (QoIs) included iso-percentiles of dosimetric exposure quantities, current intensities and densities, charge injection, off-target stimulation predictors, tissue heating, as well as tissue damage predictors – all as a function of varying degrees of fiber recruitment. The pipeline is implemented on the o2S2PARC platform for open and FAIR computational modeling [2] using modeling functionalities from Sim4Life [3]. Variability was quantified through iteration over histological samples from different subjects, multiple sources of numerical uncertainty were quantified, and model parameter uncertainties (e.g., tissue properties, fiber statistics) were propagated using advanced surrogate modeling methodologies.

Results

E-field thresholds were compared to safety guidelines [4], temperature increases to FDA limits [5], and the commonly applied (though questionably relevant) Shannon’s Criteria [6] was evaluated. The surrogate model-based uncertainty propagation permitted to shed light on complex correlations between model parameters and QoIs, fully accounting for non-linear dependencies and multi-factor interactions, and revealing novel mechanistic insights.
Discussion
The fully automatized pipeline enables quantitative safety assessment for a wide variety neural interfaces for bioelectronic medicine. It supports electrode design optimization towards improved safety and effectivity, and the identification of safe therapeutic windows. The systematic uncertainty analysis using advanced surrogate-model-based techniques illustrates the value of o2S2PARC intelligent metamodeling framework and scalable cloud resources for exploring large parameter spaces. In conclusion, carefully executed, regulatory-gradein silicosafety assessment is a powerful tool for accelerating medical device innovation.

Figure 1. Figure 1. (a) Safety assessment pipeline on o2S2PARC; (b) histology-based nerve model-generation and population with electrophysiological fiber models; (c) visualization of selected dosimetric and thermal distributions; (d) cross sections through the surrogate models with associated interpolation uncertainty; uncertainty propagation of EM and thermal tissue properties through QoI surrogate models.
Acknowledgements
“This research is supported by the NIH Common Fund’s SPARC program under award3OT3OD025348-01S8”
References
[1] NIH SPARC program, USA.https://commonfund.nih.gov/sparc
[2] Neufeld E. et al. 2020. SPARC’s Open Online Simulation Platform: o2S2PARC. FASEB J 34(S1).
[3] Sim4Life, ZMT Zurich MedTech AG, Zurich, Switzerland.
[4] ICNIRP. 2010. Guidelines for exposure to time-varying EM fields (1 Hz–100 kHz). Health Phys 99(6):818-36.
[5] FDA guidance on thermal effects:https://www.fda.gov.
[6] Shannon RV. 1992. Safe levels for electrical stimulation. IEEE Trans Biomed Eng 39(4):424-6.

Speakers

Antonino Cassara

Esra Neufeld

Javier Garcia Ordonez

Werner Van Geit

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P040: Modeling Transcranial Magnetic Stimulation: From Exposure in Personalized Head Models to Single Cell and Brain Network Dynamics Responses

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P040 Modeling Transcranial Magnetic Stimulation: From Exposure in Personalized Head Models to Single Cell and Brain Network Dynamics Responses

Antonino M. Cassara*1, Serena Santanche2, Micol Colella2, Chiara Billi1, Micaela Liberti2, Esra Neufeld1

1Foundation for Research on Information Technologies in Society (IT'IS), Zurich, Switzerland
2Universita La Sapienza, Rome, Italy

*Email: cassara@itis.swiss
Introduction

To facilitate the design and interpretation of human studies involving Transcranial Magnetic Stimulation (TMS), we developed a cloud-based and web-accessible computational framework that enables the execution of subject-specific virtual TMS experiments towards assessing and optimizing safety and efficacy. It also facilitates the formulation of testable hypotheses regarding stimulation mechanisms across various temporal and spatial scales, including mechanisms by which induced electric fields (E-fields) interact with individual neurons and high level brain network dynamics.

Methods
The framework extends a previously established pipeline [1] for non-invasive brain stimulation modeling, which combined image-based generation of detailed head models (personalized anatomy and tissue properties), personalized electromagnetic simulations (exposure and lead-fields for virtual EEG), image-based brain network model construction (mean field), and dynamic functional connectivity assessment. The current work extends this pipeline with a) TMS coil modeling and positioning, b) neuron polarization and stimulation probability mapping based on statistical sub- and supra-threshold responses of morphologically-detailed cortical neuron populations, and c) derived coupling terms for the assessment of TMS impact on network dynamics.

Results
The pipeline was employed to investigate TMS stimulation mechanisms at the single-cell and the network dynamics level. Key findings include: the dielectric contrast between gray and white matter is insufficient to directly induce spiking; mapping functions for population- and orientation-dependent threshold E-fields probabilities have been established for various pulse shapes (Figure 1), offering insights into the stimulability of different neuronal populations; electrophysiology-based activation maps have been generated for simplified models of commercial TMS coils under relevant stimulation conditions. Model validation is ongoing.

Discussion
Our pipeline extends prior modeling work [1-3] to provide a customizable framework for investigating TMS mechanisms and designing virtual clinical trials. Probability maps link dosimetric exposure predictions with electrophysiological responses that in turn modulate brain network dynamics. The pipeline serves to shed light on interaction mechanisms on to help design superior stimulation paradigms, tuned towards optimization electrophysiological response, with improve selectivity, effectivity, and safety.

Figure 1. Figure 1. (a) Illustration of the segmented head model, with 40 tissues; (b) example of user-defined TMS coils; (c) final model, featuring the optimally placed TMS coil; (d) neuronal population, orientation and pulse-specific threshold E-fields; (d) spiking threshold maps for several neuronal populations.
Acknowledgements
“This research is supported by the NIH Common Fund’s SPARC program under award3OT3OD025348-01S8”
References
[1] Karimi, F., et al. (2025). Precision non-invasive brain stimulation: an in silico pipeline for personalized control of brain dynamics. J. Neural Eng., 10.1088/1741-2552/adb88f.
[2] Aberra, A.S., et al. (2020). Simulation of TMS in head model with morphologically-realistic cortical neurons. Brain Stimul., 13(1):175-189.
[3] Jansen, B.H., Rit, V.G. (1995). EEG and VEP generation in a model of coupled cortical columns. Biol. Cybern., 73:357–366.

Speakers

Antonino Cassara

Esra Neufeld

Micol Colella

Serena Santanche

Chiara Billi

Micaela Liberti

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P041: Balanced inhibition allows for robust learning of input-output associations in feedforward networks with Hebbian plasticity

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P041 Balanced inhibition allows for robust learning of input-output associations in feedforward networks with Hebbian plasticity

Gloria Cecchini*1, Alex Roxin1

1Centre de Recerca Matemàtica, Barcelona, Spain

*Email: gcecchini@crm.cat

Introduction

In neural networks, post-synaptic activity depends on multiple pre-synaptic inputs. Hebbian plasticity allows sensory inputs to be associated with internal states, as seen in the CA1 region of the hippocampus. By modifying synaptic weights, Hebbian rules enable sensory inputs to elicit correlated outputs, allowing for efficient memory storage. When input and output patterns are uncorrelated, numerous associations can be encoded. However, if output patterns weakly correlate with input patterns, Hebbian learning reinforces shared synapses across patterns, leading to reduced network flexibility and impaired associative learning.

Methods
We analyzed the effects of Hebbian plasticity in a feedforward network model where input-output correlations emerge due to intrinsic connectivity. Using numerical simulations, we examined how weak correlations between inputs and outputs shape synaptic weight dynamics over time. We then introduced a balanced inhibition mechanism, inspired by in-vivo cortical circuits [1], to assess its impact on synaptic weight distribution and the network’s ability to store diverse associations. Network performance was evaluated by measuring output pattern variability.

Results
Our results show that when weak correlations exist between input and output patterns, Hebbian learning selectively strengthens synapses shared across patterns. This reinforcement leads to a rigid network state, where outputs become highly correlated over time. Consequently, the network loses the ability to store multiple distinct associations, significantly reducing its learning capacity. However, introducing balanced inhibition prevents the over-strengthening of shared synapses, allowing output patterns to remain distinct and ensuring a more flexible associative learning process.

Discussion
These findings highlight a fundamental limitation of Hebbian learning in feedforward networks when input-output correlations exist. Without a regulatory mechanism, the network structure becomes overly rigid, preventing effective storage of new associations. Balanced inhibition emerges as a simple yet effective strategy to mitigate this issue, preserving learning flexibility by counteracting correlation-driven synaptic reinforcement. Our study underscores the critical role of inhibition in biological neural circuits, offering insights into how the brain maintains efficient and adaptive information processing.

Acknowledgements
This project has received funding from Proyectos De Generación De Conocimiento 2021 (PID2021-124702OB-I00). This work is supported by the Spanish State Research Agency, through the Severo Ochoa and Maria de Maeztu Program for Centers and Units of Excellence in R&D (CEX2020-001084-M). We thank CERCA Programme/Generalitat de Catalunya for institutional support.
References
1. Bilal Haider, Alvaro Duque, Andrea R. Hasenstaub, David A. McCormick (2006) Neocortical Network Activity In Vivo Is Generated through a Dynamic Balance of Excitation and Inhibition. Journal of Neuroscience, 26 (17) 4535-4545; https://10.1523/JNEUROSCI.5297-05.2006

Speakers

Gloria Cecchini

Alex Roxin

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P042: Feedforward and Feedback Inhibition Flexibly Modulates Theta-Gamma Cross-Frequency Interactions in Neural Circuits

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P042 Feedforward and Feedback Inhibition Flexibly Modulates Theta-Gamma Cross-Frequency Interactions in Neural Circuits

Dimitrios Chalkiadakis*1,2, Jaime Sánchez-Claros1, Víctor J López-Madrona3, Santiago Canals2, Claudio R. Mirasso1

1Instituto de Física Interdisciplinar y Sistemas Complejos (IFISC),Consejo Superior de Investigaciones Científicas (CSIC) - Universitat de les Illes Baleares(UIB), Palma de Mallorca, Spain
2Instituto de Neurociencias, Consejo Superior de Investigaciones Científicas (CSIC) - Universidad Miguel Hernández (UMH), Sant Joan d’Alacant, Spain
3Institut de Neurosciences des Systèmes, Aix Marseille Univ -Inserm,Marseille, France

*Email:dimitrios@ifisc.uib-csic.es
Introduction
Brain rhythms are essential for coordinating neuronal activity. Cross-frequency coupling (CFC), particularly between theta (~8 Hz) and gamma (~30 Hz) rhythms, is critical for memory formation [1]. Traditionally, CFC was attributed to slow oscillations modulating faster activity at specific phases. However, metrics such as Cross-Frequency Directionality (CFD) have revealed bidirectional interactions, with both slow-to-fast and fast-to-slow influences [1, 2]. Here, we introduce a computational circuit model that flexibly exhibits both directionality interactions based on the balance of inhibitory feedforward and feedback motifs. Our framework is supported by electrophysiology measurements in the rat’s hippocampus.

MethodsWe analyzed two motifs based on variations of the (Pyramidal) Interneuron Network Gamma (PING/ING) models, both of which generate gamma rhythms through interactions between pyramidal cells (PCs) and inhibitory basket cells (BCs). An external theta drive modulatedthe network’s activity, inducing cross-frequency interactions (see Fig. 1). Somatic transmembrane currents were computed for cross-frequency dynamics analysis.Our model was validated using the experimental dataset presented in [1], which includes a detailed analysis of pathway-specific field potentials reflecting the activity of Entorhinal Cortex layer III (ECIII) projections to the hippocampal CA1 area in rats navigating both familiar and novel environments.

Results
Our analysis revealed that in θ-ING motifs, feedforward recruitment of BCs drives gamma-to-theta directionality (CFD<0), while in θ-PING motifs, feedback inhibition favors theta-to-gamma directionality (CFD>0, Fig. 1b-iii vs 1c-iii). In combined motifs, varying synaptic strengths within realistic ranges, we found smooth transitions between directionalities (Fig. 1d). Experimental data validated our framework, as behavioral conditions modulated CFD and gamma frequency in line with our model predictions (Fig. 1e). Finally, by evaluating each motif’s capacity to integrate distinct inputs impinging at different sites of the PC dendritic tree, we report their differential role in prioritizing transmission across different information channels.

Discussion
Our framework suggests that feedforward/feedback inhibitory balance regulates the directionality of theta-gamma interactions. Notably, θ-ING/θ-PING modes exist along a continuum rather than as distinct alternatives. In our model, CFD analysis identified transitions between functional modes, aligning with experimental observations across different behavioral states.
We further showed that a feedback-shifted balance promotes strong afferent-driven cross-frequency rhythmicity, while a feedforward-shifted motif broadens encoding windows,favoring parallel pathway transmission. Thus, dynamic CFD measures may reflect predominant inhibitory motifs and flexible prioritization of functional connectivity pathways.

Figure 1. Figure 1. (a) Motifs’ connections with dashed lines differentiating θ-ING (purple) from θ-PING (blue). (b, c) Cross-frequency interactions in θ-ING and θ-PING. (i) Transmembrane currents (gray) with PC/BC spikes in blue/orange. (ii) Cross-frequency coupling. (iii) CFD. (d) Mixed θ-ING/θ-PING motifs show CFD changes inversely to peak γ. (e) Experiments confirm the CFD–γ peak relationship of (d).
AcknowledgementsD. C., J. C. and C. M. acknowledge support from the Spanish Ministerio de Ciencia, Innovación y Universidades through projects PID2021-128158NB-C22 and María de Maeztu CEX2021-001164-M. D. C. and S. C. acknowledge support from the Spanish Ministerio de Ciencia, Innovación y Universidades through projects PID2021-128158NB-C21 and Severo Ochoa CEX2021-001165-S
References
[1] López-Madrona, V. J., Pérez-Montoyo, E., Álvarez-Salvado, E., Moratal, D., Herreras, O., Pereda, E., … Canals, S. (2020). Different Theta Frameworks Coexist in the Rat Hippocampus and Are Coordinated during Memory-Guided and Novelty Tasks.eLife,9, e57313. doi:10.7554/eLife.57313
[2] Jiang, H., Bahramisharif, A., Van Gerven, M. A. J., & Jensen, O. (2015). Measuring Directionality between Neuronal Oscillations of Different Frequencies.NeuroImage,118, 359–367. doi:10.1016/j.neuroimage.2015.05.044

Speakers

Dimitrios Chalkiadakis

Jaime Sánchez-Claros

Santiago Canals

Claudio Mirasso

Víctor José López Madrona

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P043: LEC Ensemble Analysis Reveals Context-Guided Odor Encoding via Adaptive Spatio-Temporal Representations

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P043 LEC Ensemble Analysis Reveals Context-Guided Odor Encoding via Adaptive Spatio-Temporal Representations

Yuxi Chen*1, Noga Mudrik*2, James J. Knierim1, Adam S. Charles2

1Department of Neuroscience, Mind/Brain Institute, Johns Hopkins University, Baltimore, USA
2Department of Biomedical Engineering, Kavli NDI, Center of Imaging Science, Johns Hopkins University, Baltimore, USA
*Email: ychen315@jhu.edu; nmudrik1@jhu.edu

Introduction.The lateral entorhinal cortex (LEC) has rich associative connections in the rat cortex, linking the hippocampus and neocortex. It encodes spatial context and develops odor selectivity [2,3]. A key question is how LEC enables context-odor integration. Using Neuropixels, we recorded LEC activity while rats performed an odor-context task that included identifying an odor and selecting the corresponding reward port, with the reward-port location switching between two contexts defined by the box the rat occupied (Fig. A). We further developed an ensemble-identification method to reveal how hidden LEC ensembles support context-odor integration via adaptive representation pre- vs. post-training.Methods.We recorded rats’ LEC both before and after learning odor-context associations (“day 1” vs. “day N”). Each day had 6 sessions, with the rat alternately placed in one of two boxes that provide context through unique cues (Fig. A). Spike sorting was done with Kilosort4, and firing rate (FR) was estimated via Gaussian convolution, producing a multi-label tensor dataset (Fig. B). We developed a graph-driven ensemble method that extends [4] to multi-class data, identifies state-dependent ensemble composition (A) adjustments (Fig. D), and captures per-trial temporal variability in ensemble traces (φ). We tested the ensembles’ encoding of odor vs. box via 5-fold cross-validation logistic regression.Results.Neurons vary by odor, box, or both (Fig. C), suggesting ensembles. We found ensembles with session adjustments (Fig. H), that temporally encode box (Fig. E). On day 1, ensemble traces differentiate boxes over time; while on day N, boxes are more separated at trial start. Session-trace averages under fixed box (Fig. F) show more variability on day 1 compared to day N, which featured consistency of same-box sessions. Odor encoding is less apparent, and, on day N, is primarily revealed under a fixed box (Fig. G). Odor prediction accuracy improved when conditioned on box (Fig. I), with a larger improvement on day N. Odor feature importance shows that conditioning on the box shifts encoding timing, with later points more important under fixed-box.Discussion.We identified session-adjusting ensembles that capture box encoding, with earlier encoding on day N. On day 1, rats show distinct representations for same-box sessions, suggesting session-by-session encoding, while on day N, consistent same-box session activations suggest box recognition. We hypothesize that post-training, rats first identify the box, which opens an 'odor-integration gate'. This aligns with improved odor-encoding accuracy and the shift in odor timing importance when conditioned on the box compared to marginalized. Our findings suggest that over training, the LEC develops a hierarchical mechanism for context-odor integration that starts with early context identification, followed by box-conditioned odor integration.

Figure 1. A: Experiment. B: Multi-class data across box-odor-sessions. C: Single neuron traces. D: Ensembles-adjusting approach leveraging [4]. E–G: Ensemble traces by box (E), session (F), and odor (G, left: marginalizing, right: conditioning on box). H: Two day-N ensembles adjusting by session. I: Odor prediction confusion matrices ± box conditioning. J: Ensemble/time point importance for odor encoding.
Acknowledgements

Y.C. and J.J.K. were fundedbyNIA grant P01 AG009973.N.M. was funded by the Kavli Foundation Neurodiscovery Award and as a Kavli Fellow of Johns Hopkins Kavli NDI. A.S.C was supported by NSF CAREER Award 2340338 and a Johns Hopkins Bridge Grant.
References
●
Bota, M., Sporns, O., & Swanson, L. W. (2015). Architecture of the cerebral cortical association connectome underlying cognition.PNAS.
●
Igarashi, K. M., et al. (2014). Coordination of entorhinal–hippocampal ensemble activity during associative learning.Nature.
●
Tsao, A., Sugar, J., Lu, L., Wang, C., Knierim, J. J., Moser, M. B., & Moser, E. I. (2018). Integrating time from experience in the lateral entorhinal cortex.Nature.
●
Mudrik, N., Mishne, G., & Charles, A. S. (2024). SiBBlInGS: Similarity-driven Building-Block Inference using Graphs across States. ICML.

Speakers

Yuxi Chen

Noga Mudrik

James Knierim

Adam Charles

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P044: Uncertainty-Calibrated Network Initialization via Pretraining with Random Noise

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P044 Uncertainty-Calibrated Network Initialization via Pretraining with Random Noise

Jeonghwan Cheon*1, Se-Bum Paik1

1Department of Brain and Cognitive Sciences, Korea Advanced Institute of Science and Technology, Daejeon, Republic of Korea

*Email: jeonghwan518@kaist.ac.kr

Uncertainty calibration — the ability to estimate predictive confidence that reflects the actual accuracy — is essential to real-world decision-making. Human cognition involves metacognitive processes, allowing us to assess uncertainty and distinguish between what we know and what we do not know. In contrast, current machine learning models often struggle to properly calibrate their confidence, even though they have achieved high accuracy in various task domains [1]. This miscalibration presents a significant challenge in real-world applications, such as autonomous driving or medical diagnosis, where incorrect decisions can have critical consequences. Although post-processing techniques have been used to address calibration issues, they require additional computational steps to obtain reliable confidence estimates. In this study, we show that random initialization — a common practice in deep learning — is a fundamental cause of miscalibration. We found that randomly initialized, untrained networks exhibit excessively high confidence despite lacking meaningful knowledge. This miscalibration at the initial stage prevents the alignment of confidence with actual accuracy as the network learns from data. To address this issue, we draw inspiration from the developmental brain, which is initialized through spontaneous neural activity even before receiving sensory inputs [2]. By mimicking this process, we pretrain neural networks with random noise [3] and demonstrate that this simple approach resolves the overconfidence issue, bringing initial confidence levels to near chance. This pre-calibration through random noise pretraining enables optimal calibration by aligning confidence levels with actual accuracy during subsequent data training. As a result, networks pretrained with random noise achieve significantly lower calibration errors compared to those trained solely with data. We also confirmed that this method generalizes well across different conditions, regardless of dataset size or network complexity. Notably, these pre-calibrated networks consistently identify “unknown data” by showing low confidence for outlier inputs. Our findings present a key solution for calibrating uncertainty in both in-distribution and out-of-distribution scenarios without the need for post-processing. This provides a fundamental approach to addressing miscalibration issues in artificial intelligence and may offer insights into the biological development of metacognition.
Acknowledgements
This work was supported by the National Research Foundation of Korea (NRF) grants (NRF-2022R1A2C3008991 to S.P.) and by the Singularity Professor Research Project of KAIST (to S.P.).
References
● Guo, C., Pleiss, G., Sun, Y., & Weinberger, K. Q. (2017). On calibration of modern neural networks. InInternational Conference on Machine Learning(pp. 1321-1330). PMLR.
● Martini, F. J., Guillamón-Vivancos, T., Moreno-Juan, V., Valdeolmillos, M., & López-Bendito, G. (2021). Spontaneous activity in developing thalamic and cortical sensory networks.Neuron,109(16), 2519-2534.
● Cheon, J., Lee, S. W., & Paik, S. B. (2024). Pretraining with random noise for fast and robust learning without weight transport.Advances in Neural Information Processing Systems,37, 13748-13768.

Speakers

Jeonghwan Cheon

Se-Bum Paik

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P045: Cortical Microcircuit Modeling for Concurrent EEG-fMRI Recordings

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P045 Cortical Microcircuit Modeling for Concurrent EEG-fMRI Recordings

Shih-Cheng Chien*1, Stanislav Jiříček1,2,3, Thomas Knösche4, Jaroslav Hlinka1,2, Helmut Schmidt1

1Institute of Computer Science of the Czech Academy of Sciences, Prague, Czech Republic
2National Institute of Mental Health, Klecany, Czech Republic
3Department of Cybernetics, Faculty of Electrical Engineering, Czech Technical University in Prague, Prague, Czech Republic
4Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig, Germany

*Email:chien@cs.cas.cz
Introduction

EEG and fMRI are widely used noninvasive methods for human brain imaging. Concurrent EEG-fMRI recordings help answer fundamental questions about the functional roles of EEG rhythms, their origin, and their relationship with the BOLD signal. Given the fact that both EEG and BOLD signals predominantly originate from postsynaptic potentials (PSPs) [1,2], and considering that distinct inhibitory neuron types influence EEG rhythms differently [3] and possess varied neurovascular coupling properties [4], a cortical microcircuit model incorporating multiple inhibitory neuron types would offer a promising framework for investigating local neural dynamics underlying EEG rhythms and their relationship with BOLD signals.

Methods
We developed a cortical microcircuit model that incorporates excitatory (E) and inhibitory (PV, SOM, and VIP) populations across cortical layers (L2/3, L4, L5, and L6) with realistic configurations, including connection probabilities, synaptic strengths, neuronal densities, and firing rate functions for each neuron type. The model receives three types of external inputs: (1) lateral input, (2) modulatory input, and (3) thalamic input. We characterized the spectral properties of EEG rhythms across a range of external inputs, explored EEG-BOLD correlations under constant and varying input conditions, and analyzed how neuronal populations contribute to EEG rhythms and the EEG-BOLD correlation.
Results
The model generates EEG rhythms, with increased power in the alpha (8-12 Hz), beta (13-30 Hz), and gamma bands (30-50 Hz) at low modulatory input and increased delta (0.5-4 Hz) and theta (4-7 Hz) powers at high modulatory input. We found low-frequency EEG activity (from delta to low beta band) was driven more strongly by infragranular than supragranular populations. Conversely, supragranular populations drive high-frequency EEG activity (high beta and gamma band) more intensely. As to EEG-BOLD correlations, we found that alpha-BOLD correlation is almost exclusively driven by fluctuations (i.e., standard deviation of firing rates) in infragranular populations, with little contribution from the supragranular layer.

Discussion
Our cortical microcircuit model generates EEG rhythms based on a generic mechanism involving the nonlinear amplification and filtering of synaptic noise. Our investigation focused on different forms of long-range external input, which targets distinct neuronal populations. The model could be used to help design optimal stimulation protocols for various applications, including the effect of specific neuronal populations on EEG and BOLD.

Acknowledgements
The publication was supported by a Lumina-Quaeruntur fellowship (LQ100302301) by the Czech Academy of Sciences (awarded to HS) and ERDF-Project Brain Dynamics, No. CZ.02.01.01/00/22_008/0004643. We acknowledge the core facility MAFIL supported by the Czech-BioImaging large RI project (LM2018129 funded by MEYS CR) for their support in obtaining scientific data presented in this work.
References
[1]https://doi.org/10.1016/j.brainresrev.2009.12.004
[2]https://doi.org/10.1016/j.cub.2018.11.052
[3]https://doi.org/10.1016/j.tins.2003.09.016
[4]https://doi.org/10.1523/JNEUROSCI.3065-04.2004

Speakers

Helmut Schmidt

Scientific researcher, Institute of Computer Science, Czech Academy of Sciences

Jaroslav Hlinka

Senior researcher, Institute of Computer Science of the Czech Academy of Sciences

Currently I am leading the COBRA working group and also serve as the Head of the Department of Complex Systems and as the Chair of the Council of the Institute of Computer Science of the Czech Academy of Sciences.Brief bio After obtaining master degrees in Psychology from Charles University (2005) and in Mathematics from Czech Technical University (2006), I went on the quest of applying mathematics in helping to understand the complex activity of human bra... Read More →

Shih-Cheng Chien

Stanislav Jiricek

Thomas R. Knösche

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P046: Model Parameter Estimation for TMS-induced MEPs

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P046 Model Parameter Estimation for TMS-induced MEPs

Shih-Cheng Chien*1, Christian Röse2, Peng Wang2,3, Helmut Schmidt1, Jaroslav Hlinka1,4, Thomas R. Knösche2, Konstantin Weise2

1Institute of Computer Science of the Czech Academy of Sciences, Prague, Czech Republic
2Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig, Germany
3Institute of Psychology, University of Greifswald, Greifswald, Germany
4National Institute of Mental Health, Klecany, Czech Republic

*Email:chien@cs.cas.cz

Introduction

TMS-induced MEPs are widely utilized in both basic research and clinical practice. The MEP parameters, such as input-output (I/O) curves, often exhibit significant variability across individuals, both in healthy populations and in patients. Understanding the sources of this variability is critical for improving the precision of motor-related diagnoses. Previously, we developed a biologically inspired model capable of reproducing MEP waveforms. In this study, we apply model fitting to an open MEP dataset of ten healthy participants [1] and investigate the distribution of model parameters underlying the variability of I/O curves.

Methods
The model incorporates the descending motor pathways from the spinal cord to the hand muscles, with synthetic D- and I-waves serving as inputs. The spinal cord component consists of 100 conductance-based leaky integrate-and-fire alpha motor neurons (aMNs), which interact with a population of Renshaw cells (RCs) that function as a common inhibitory pool. The aMNs are connected to 100 motor units in the hand muscle component. Each motor unit generates a motor unit action potential (MUAP) in response to spikes from its corresponding aMN. The simulated MEP is computed as the sum of these time-shifted MUAPs.
Results
The resting motor threshold (RMT) across individuals in the dataset was 41.3 ± 6.0% of the maximum stimulator output (MSO). Peak latencies showed no significant variation with MEP peak-to-peak amplitude. Fitting the model to individual MEP waveforms provided insights into the neuronal interactions underlying MEP generation. The DI-waves, after convolution with synaptic kernels (AMPA and NMDA), produced sustained inputs to the αMNs. Renshaw cells played a critical role in suppressing excessive spikes, particularly under high TMS intensities, preventing excessive oscillations in the MEP waveform.
Discussion
We employed a computationally efficient and biologically plausible model to explain the variability in individual TMS-induced MEPs. The fitting procedure relied on synthesizing common DI-waves for healthy participants, which may introduce additional errors in parameter estimation. Future work will validate this approach using patient data, where individual DI-waves are available, to improve accuracy and robustness in parameter fitting.

Acknowledgements

The publication was supported by a Lumina-Quaeruntur fellowship (LQ100302301) by the Czech Academy of Sciences (awarded to HS) and ERDF-Project Brain Dynamics, No. CZ.02.01.01/00/22_008/0004643.
References
[1]https://doi.org/10.1016/j.brs.2022.06.013

Speakers

Helmut Schmidt

Scientific researcher, Institute of Computer Science, Czech Academy of Sciences

Jaroslav Hlinka

Senior researcher, Institute of Computer Science of the Czech Academy of Sciences

Shih-Cheng Chien

Thomas R. Knösche

Peng Wang

Konstantin Weise

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P047: Graph Analysis of EEG Functional Connectivity during Lie Detection

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P047 Graph Analysis of EEG Functional Connectivity during Lie Detection

Yun-jeong Cho1, Hoon-hee Kim*2

1 Department of Data Engineering, Pukyong National University, Busan, South Korea
2Department of Computer Engineering and Artificial Intelligence, Pukyong National University, Busan, South Korea

*Email: h2kim@pknu.ac.kr

Introduction

Lie detection is an important research topic in various fields, including psychology, forensic science, and neuroscience, as it involves complex processes. By measuring the brain's functional connectivity using EEG data and calculating the graph-theoretical metrics (e.g., the average clustering coefficient), it is possible to quantitatively assess the changes in brain network dynamics between lie and truth conditions [1]. In this study, we aimed to analyze the overall brain connectivity differences between lie and truth conditions by computing inter-channel coherence and network metrics within a specific frequency band during the answer phase.
Methods
Among 12 subjects, participants were divided into two groups —those who consistently lied and those who consistently told the truth. After excluding two lie subjects, each group comprised five subjects. EEG data were recorded for 15 seconds while subjects answered a specific question with only the first 3 seconds after answer onset analyzed. Inter-channel coherence [2] was computed in the high-frequency range, focusing on the beta band activated during lying. A functional connectivity (FC) matrix was constructed by applying a threshold, and key metrics —such as the average clustering coefficient and global efficiency were calculated. Statistical validation was performed using t-tests and Mann-Whitney U tests.
Results
Overall, significant differences in brain network metrics were observed between the lie and truth conditions (Fig. 1). In particular, the subjects in the lie group, the average clustering coefficient was found to increase significantly than in the subjects in the truth group. Statistical analyses confirmed that these differences were significant, with a larger than expected effect size, suggesting that overall brain connectivity is altered when individuals lie. These findings support the notion that the complex cognitive processes involved in lying may lead to changes in the brain’s network organization.
Discussion
This study compared the overall brain network changes between lie and truth conditions using the average clustering coefficient computed for each subject. The results showed that the lie condition exhibited increased global brain connectivity, suggesting an additional cognitive load during lying. However, using subject-level averages limits the ability to directly assess local connectivity changes in specific brain regions, and caution is warranted in interpretation due to the small sample size. Future research should include a larger number of subjects and incorporate various network metrics, such as inter-channel analyses, to more precisely evaluate brain connectivity changes.

Figure 1. Fig 1. Topographic maps of the average clustering coefficient comparing lie (left) and truth (right) groups during the answer phase. Increased clustering (dark red) in the lie condition indicates significantly greater overall brain connectivity compared to the truth condition.
Acknowledgements
This study was supported by the National Police Agency and the Ministry of Science, ICT & Future Planning (2024-SCPO-B-0130), the National Research Foundation of Korea grant funded by the Korea government (RS-2023-00242528), the National Program for Excellence in SW, supervised by the IITP(Institute of Information & communications Technology Planing & Evaluation) in 2025(2024-0-00018)
References
1. Gao J, Gu L, Min X et al. (2022). Brain Fingerprinting and Lie Detection: A Study of Dynamic Functional Connectivity Patterns of Dedeption Using EEG Phase Synchrony Analysis. IEEE Journal of Biomedical and Health Informatics, 26(2), 600-613.https://doi.org/10.1109/jbhi.2021.3095415
2. Bowyer S. (2016). Coherence a measure of the brain networks: past and present. Neuropsychiatric Electrophysiology, 2(1).https://doi.org/10.1186/s40810-015-0015-7

Speakers

Yun Jeong Cho

Hoon-hee Kim

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P048: From Density to Void: Why Brain Networks Fail to Reveal Complex Higher-Order Structures

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P048 From Density to Void: Why Brain Networks Fail to Reveal Complex Higher-Order Structures

Moo K. Chung*1,Anass B El-Yaagoubi2, Anqi Qiu3, Hernando Ombao2

1Department of Biostatistics and Medical informatics, University of Wisconsin, Madison, USA
2Statistics Program, King Abdullah University of Science and Technology, Thuwal, Saudi Arabia
3Department of Health Technology and Informatics, Hong Kong, China

*Email:mkchung@wisc.edu

Introduction

In brain network analysis using resting-state fMRI, there is growing interest in modeling higher-order interactions—beyond simple pairwise connectivity—using persistent homology [1]. Despite the promise of these advanced tools, robust and consistently observed time-evolving higher-order interactions remain elusive. In this study, we examine why conventional analyses often fail to reveal complex higher-order structures—such as interactions involving four or five or more nodes —and explore whether higher-order interactions truly exist in functional brain networks.

Methods

We apply persistent homology to analyze correlation networks over a range of thresholds h. A simplicial complex is constructed from the connectivity matrix c(i,j) where nodes (0-simplices) represent individual time series and edges (1-simplices) are included if c(i,j) > h. For triangles (2-simplices), a simplex is formed if all three pairwise connections among a triplet of nodes exceed the threshold h. Higher-order simplices are defined analogously. We then examine the consistency of these higher-order topological features across time and subjects by quantifying the probability of overlap in the persistent features.

Results

Our preliminary analysis based on rs-fMRI of 400 subjects reveals that correlation networks tend to yield either nearly complete graphs or highly fragmented structures, neither of which exhibit robust higher-order topological features. As the number of nodes involved in an interaction increases, the probability that multiple brain regions activate simultaneously decays exponentially, as observed in both empirical data and theoretical models. These findings indicate that resting-state fMRI predominantly reflects pairwise interactions, with only infrequent occurrences of three-node interactions. Nonetheless, even these predominant pairwise interactions are highly intricate, giving rise to complex network dynamics characterized by lower-dimensional topological profiles such as 0D (connected components) and 1D (cycles) features [2].

Discussion
Our results indicate that conventional connectivity analyses are limited in detecting robust higher-order interactions, as they often yield networks that are either overly dense or fragmented, masking subtle connectivity patterns. Alternative metrics, such as mutual information or entropy, may better capture the nonlinear, multiscale dependencies among brain regions [3]. Notably, higher-order interactions are not exclusively defined by multi-node connectivity; even pairwise interactions can become highly complex when organized into cycles or spiral patterns over time. Future work should integrate these alternative measures with persistent homology to reveal hidden connectivity patterns, ultimately enhancing our understanding of functional brain organization.

Figure 1. Left: Graph representation of pairwise interactions between nodes in a brain network. Right: Higher-order interactions depicted with colored simplices—yellow for 3-node (triangle) interactions and blue for 4-node (tetrahedron) interactions.
Acknowledgements
NIH grants EB028753, MH133614 and NSF grant MDS-201077

References
[1] El-Yaagoubi, A.B., Chung, M.K., Ombao. H. (2023). Topological data analysis for multivariate time series data.Entropy,25(11), 1509.
[2]Chung, M.K., Ramos, C.G., De Paiva, F.B., Mathis, J., Prabhakaran, V., Nair, V.A., Meyerand, M.E., Hermann, B.P., Binder, J.R. and Struck, A.F., 2023. Unified topological inference for brain networks in temporal lobe epilepsy using the Wasserstein distance.NeuroImage,284, p.120436.
[3] Li, Q., Steeg, G. V., Yu, S., Malo, J. (2022). Functional connectome of the human brain with total correlation.Entropy,24(12), 1725.

Speakers

Moo Chung

Anass El-Yaagoubi

Anqi Qiu

Hernando Ombao

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P049: Planning and hierarchical behaviors in homeostatic optimal control

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P049 Planning and hierarchical behaviors in homeostatic optimal control

Simone Ciceri*1,2, Atilla-Botond Kelemen1,3, Henning Sprekeler1,3,4

1Modelling of Cognitive Processes, Technical University of Berlin, Berlin, Germany
2Charité–Universitätsmedizin Berlin, Einstein Center for Neurosciences Berlin, Berlin, Germany
3Bernstein Center for Computational Neuroscience, Berlin, Germany
4Science of Intelligence, Research Cluster of Excellence, Berlin, Germany

*Email:simone.ciceri@tu-berlin.de

Introduction
Animal survival depends on the ability to maintain the stability of a set of internal variables, such as nutrient levels or water balance. This internal regulation, known as homeostasis, often requires the acquisition of resources via interactions with the external environment [1]. We reasoned that competition among multiple homeostatic needs combined with a rich environment may be sufficient to explain a wide range of complex behaviors. To test this hypothesis, we developed a control-theoretic problem setting for an agent that aims to preserve homeostasis of multiple internal variables while foraging in environments with distributed resources.

Methods
We model a synthetic agent that actively forages to minimize deviations of its internal variables from their respective set points, which reflect its individual demands. These variables gradually decay over time but can be replenished by collecting resources from the environment. The resources are distributed around the environment to generate competition among different needs. We study the foraging behavior that results from minimizing a cost function that combines homeostatic errors and motion costs. In simple 1D environments, we obtain optimal behavioral policies using optimal control methods. In 2D settings, we parametrize the policies with artificial neural networks that are optimized using evolutionary algorithms.

Results
We show that internal homeostasis can generate a rich repertoire of behaviors that depend on both the structure of the environment and internal demands. First, when resources are sparse the agent displays planning strategies, such as stocking up on one variable before foraging for others. Second, agent behaviors can be decomposed into a small set of simpler policies, each of which satisfies one internal need. The agent hierarchically selects from this set of behaviors based on its internal state. Finally, optimal strategies can be highly sensitive to the agent's demands. In the same environment, we can observe sudden transitions between different behaviors when changing the set point at which the internal variables need to be maintained.

Discussion
Our model demonstrates the possible emergence of complex behavior from the simple goal of internal stability. Optimal foraging strategies are shaped by both environmental factors and internal demands, potentially accounting for the large variability often observed among individuals of the same species, even within the same environment. Our model also emphasizes how strongly the dynamics of the internal state—which are generally not accessible in behavioral experiments—are mirrored in the agent's behavior. The relevance of these findings is not confined to behavioral modeling and analysis: it is likely that the neural activity that drives animal behaviors will be similarly sensitive to the internal state of the animal.

Acknowledgements
-
References
● Woods, S. C., & Ramsay, D. S. (2007). Homeostasis: Beyond Curt Richter. Appetite, 49(2), 388-398.https://doi.org/10.1016/j.appet.2006.09.015

Speakers

Simone Ciceri

Henning Sprekeler

Atilla-Botond Kelemen

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P050: Linking biological context to models with ModelDB

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P050 Linking biological context to models with ModelDB

InessaCohen1,XinchengCai2,MengmengDu2,YitingKong2,HongyiYu2,Robert A. McDougal*1,2,3,4
1Program in Computational Biology and Biomedical Informatics, Yale University, New Haven, CT, USA
2Department of Biostatistics, Yale School of Public Health, New Haven, CT, USA
3Department of Biomedical Informatics and Data Science, YaleSchool of Medicine, New Haven, CT, USA
4Wu Tsai Institute, Yale University, New Haven, CT, USA

*Email: robert.mcdougal@yale.edu
Introduction

ModelDB (https://modeldb.science) was founded almost 30 years ago to address the challenges of reproducibility in computational neuroscience, promote code reuse, and facilitate model discovery. It has grown to hold the source code for ~1,900 published studies. Recent enhancements, presented here, have focused on expanding its model collection and improving its biological context. However, discoverability and interpretability depend on having reliable metadata for entire models and their components. To address this, we sought to use machine learning (ML) to classify ion channel subtypes based on source code, identify key predictors, and compared results to those from a large language model (LLM).
Methods
We applied manual and automatic techniques to increase the biological context displayed when exploring ModelDB as well as increased the visibility of existing data. Network model properties and some file-level ion channel types were manually annotated. Biology-focused explanations of model files were generated automatically by an LLM. Features were extracted using a rule-based approach from NEURON [1] MOD files (a common format of ion channel and receptor model components) after deduplication that ignored white space and comments. Five-fold cross-validation was used to assess ML predictions. Subsets of model code from many files and a controlled vocabulary were provided to an LLM to generate whole-model metadata which was assessed manually.
Results
We have updated the ModelDB website to support more types of models and to pair browsing models and files with biological and computational context. The ML classifier identified a number of features (state count, nonspecific currents, using common ions) as key for predicting ion channel type. It worked well for identifying broad channel types but struggled with more granular subtype identification which had few examples in our training set. Calcium-activated potassium channels were one of the best performing subtypes. ML results were compared with those from an LLM and from rule-based approaches. LLM performance on whole model metadata prediction from source code was highly dependent on the broad category of metadata.
Discussion
ModelDB has long prioritized connecting models to biology, from its days as part of the SenseLab project, where its sister-site NeuronDB [2] once gathered compartment-level channel expression data. Many model submitters now chose to contribute an “experimental motivation” when submitting new models. Biology and model code are both often unclear on what should count as “the same,” posing challenges for both manual and automated metadata assignment. Nevertheless, it is our hope that pairing code with enriched biological context will make computational models more accessible, interpretable, and reusable.

Acknowledgements
We thankRui Lifor curatingModelDBmodel network metadata.
References
1.Hines, M. L., & Carnevale, N. T. (1997). The NEURON simulation environment.Neural computation, 9(6), 1179-1209.https://doi.org/10.1162/neco.1997.9.6.1179
2.Mirsky, J. S., Nadkarni, P. M., Healy, M. D., Miller, P. L., & Shepherd, G. M. (1998). Database tools for integrating and searching membrane property data correlated with neuronal morphology.Journal of neuroscience methods, 82(1), 105-121.https://doi.org/10.1016/S0165-0270(98)00049-1

Speakers

Inessa Cohen

Xincheng Cai

Speakers

Rhiannon Cowan

Tyler S. Davis

Bornali Kundu

Ben Shofty

Shervin Rahimpour

John D. Rolston

Elliot H. Smith

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P056: A numerical simulation of neural fields on curved geometries

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P056 A numerical simulation of neural fields on curved geometries

Neekar Mohammed, David J. Chappell,Jonathan J. Crofts*
Department of Physics and Mathematics, Nottingham Trent University, Nottingham, UK

*Email: jonathan.crofts@ntu.ac.uk

Introduction
Brainwaves are crucial for information processing, storage, and sharing [1]. While a plethora of computational studies exist, the mechanisms behind their propagation remain unclear. Current models often simplify the cortex as a flat surface, ignoring its complex geometry [2]. In this study, we incorporate realistic brain geometry and connectivity into simulations to investigate how brain morphology influences wave propagation. Our goal is to leverage this approach to elucidate the relationship between increasing mammalian brain convolution, brain evolution, and its consequential impact on cognition.

Methods
To achieve efficient modelling of large-scale cortical structures, we have extended isogeometric analysis (IGA) [3], a powerful tool for physics-based engineering simulations, to the complex nonlinear integro-differential equation models found in neural field models. IGA utilises non-uniform rational B-splines (NURBS), the standard for geometry representation in computer-aided design, to approximate solutions. Specifically, we will employ isogeometric collocation (IGA-C) methods, leveraging the high accuracy of NURBS with the computational efficiency of collocation. While IGA-C has proven effective for linear integral equations in mechanics and acoustics, its application to nonlinear NFMs represents a significant advancement.
Results
To enable more realistic brain simulations, we have developed a novel IGA-C method that directly utilises point cloud data and bypasses triangular mesh generation, allowing for the solution of partial integro-differential equation models of neural activity on complex cortical-like domains. Here, we demonstrate the method's capabilities by studying both localised and traveling wave activity patterns in a two-dimensional neural field model on a torus [4]. The model offers a significant computational advantage over standard mesh-dependent methods and, more importantly, provides a crucial framework for future research into the role of cortical geometry in shaping neural activity patterns via its ability to incorporate complex geometries.
Discussion
This work presents a novel numerical procedure for integrating neural field models on arbitrary two-dimensional surfaces, enabling the study of physiologically realistic systems. This includes, for example, accurate cortical geometries and connectivity functions that capture regional heterogeneity. Future research will focus on elucidating the influence of curvature on the nucleation and propagation of travelling wave solutions on cortical geometries derived from imaging studies.

Acknowledgements
NM, DJC and JJC were supported through the Leverhulme Trust research project grant RPG-2024-114
References
1.https://doi.org/10.1038/nrn.2018.20
2.https://doi.org/10.1007/s00422-005-0574-y
3.https://doi.org/10.1016/j.cma.2004.10.008
4.https://doi.org/10.1007/s10827-018-0697-5

Speakers

Jonathan Crofts

Neekar Mohammed

David Chappell

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P057: Biologically Interpretable Machine Learning Approaches for Analyzing Neural Data

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P057 Biologically Interpretable Machine Learning Approaches for Analyzing Neural Data

Madelyn Esther C. Cruz*1,2, Daniel B. Forger1,2,3

1Department of Mathematics, University of Michigan, Ann Arbor, MI, USA
2Department of Computational Medicine and Bioinformatics, University of Michigan, Ann Arbor, MI, USA
3Michigan Center for Interdisciplinary and Applied Mathematics, University of Michigan, Ann Arbor, MI, USA

*Email:mccruz@umich.edu
Introduction

Deep neural networks (DNNs) often achieve impressive classification performance, but they operate as "black boxes,” making them challenging to interpret [1]. They may also struggle to capture the dynamics of time-series data, such as electroencephalograms (EEGs), because of their indirect handling of temporal information. To address these challenges, we explore the use of Biological Neural Networks (BNNs), machine learning models inspired by the brain’s physiology, on neuronal data. By leveraging biophysical neuron models, BNNs offer better interpretability by closely modeling neural dynamics, providing insights into how biological systems generate complex behavior.
Methods
This study applies backpropagation to networks of biophysically accurate mathematical neuron models to develop a BNN model. Specifically, we define a BNN architecture using modified versions of the Hodgkin–Huxley model [2] and integrate this within traditional neural network algorithms. These BNNs are then used to classify both EEG and non-EEG signals, generate EEG signals to predict brain states, and analyze EEG neurophysiology through model-derived parameters. We also compare the performance of our BNN architecture to those of traditional neural networks.
Results
Our BNNs demonstrate strong performance in classifying handwritten digits from the MNIST Digits Dataset, learning faster than traditional neural networks. The same BNN architecture also excels on time-series neuronal datasets, effectively distinguishing EEG recordings and power spectral densities associated with alertness vs. fatigue, varying consciousness levels, and different workloads. Additionally, we trained our BNNs to exhibit different frequencies observed in EEG recordings and found that the variability of synaptic weights and applied currents increased with the target frequency range.
Discussion
Analyzing gradients from backpropagation in BNNs reveals similarities between their learning mechanisms and Hebbian learning in the brain in terms of how synaptic weights change the loss function and how changing the weights at specific time intervals impact learning. In particular, synaptic weight updates occur only when presynaptic or postsynaptic neurons fire [3]. This results in fewer parameter changes during training compared to DNNs while still capturing temporal dynamics, leading to improved learning efficiency and interpretability. Overall, applying backpropagation to accurate ordinary differential equation models enhances neuronal data classification and interpretability while providing insights into brain learning mechanisms.

Acknowledgements
We acknowledge the following funding: ARO MURI W911NF-22-1-0223 to DBF.
References
● http://doi.org/10.1038/nature14539
● https://doi.org/10.1007/s00422-008-0263-8
● https://doi.org/10.1016/j.neucom.2014.11.022

Speakers

Madelyn Esther Cruz

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P058: Homeostatic memory engrams in mesoscopic structural connectomes

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P058 Homeostatic memory engrams in mesoscopic structural connectomes

Fabian Czappa(1, *)
Marvin Kaster (1)

Marcus Kaiser (2)
Markus Butz-Ostendorf (1,3)
Felix Wolf (1)

(1) Laboratory for Parallel Programming, Department of Computer
Science, Technical University of Darmstadt, Hochschulstraße. 10,
Darmstadt, 64285, Hesse, Germany

(2) Translational Neuroimaging, Faculty of Medicine & Health Sciences,
University of Nottingham, NG7 2RD, Nottingham, United Kingdom

(3) Translational Medicine and Clinical Pharmacology, Boehringer Ingelheim Pharma GmbH & Co. KG Birkendorfer Straße 65 88397 Biberach/ Riss, Baden-Wuerttemberg, Germany

(*)fabian.czappa@tu-darmstadt.de
Introduction

Memory engrams are defined as physical traces of memory [1]. However, their actual representation in the brain is still unknown. To shed light on the underlying mechanisms, we simulate the formation of memories in healthy human subjects based on connectomes extracted from their DT-MRI brain scans. To prepare the networks for learning, we first bring them into a state of homeostatic equilibrium, leaving their topology largely intact [2]. Once this homestatization is complete, we perform a memory experiment by stimulating groups of neurons and observing memory formation as the network changes its structure to maintain equilibrium [3]. After our "thought experiment", we can precisely locate the memory engram in the connectome.
Methods
We use the Model of Structural Plasticity (MSP) [4], which grows and retracts synaptic elements based on a homeostatic rule. When a neuron searches for a partner, it chooses one based on the number of free synaptic elements and a distance-dependent probability kernel. We augment the original kernel at longer distances, giving preference to the vicinity of established synapses. After homeostatizing the structural connectome with the augmented MSP, we select a group of concept cells (CC) from the middle temporal lobe and two groups of neurons C1 and C2 scattered outside this region. We then perform a Hebbian learning experiment, associating CC with C1. We perform our experiments using data from n=7 healthy human subjects [5].
Results
Homeostatizing the connectome brings the node-degree distribution from a power-law to a normal distribution, yet we keep many distinguishing features of the network. The (geometric) axon-length histogram, the small-worldness, and the assortativity – among others – are comparable between the scanned connectome and the homeostatized one. Furthermore, we see that we form a memory engram after picking neurons for CC, C1, and C2 and stimulating CC and C1 together. Testing with n=7 high-resolution connectomes, we see that the memory engram is located in specific brain areas such as the inferior parietal lobule (7 times), the superior temporal lobe (7 times), but only sometimes in the fusiform gyrus (4 times); see Figure 1 for details.
Discussion
For the first time, it is now possible to conduct brain simulations based on individual brain scans without parameter fitting. Using MSP-generated avatar connectomes of healthy subjects that were topologically similar to the original tractograms, our method ensured the functioning of model neurons in a physiological regime, which was the necessary precondition for the learning experiments. The proposed approach is the starting point of various testable and personalized brain simulations, from designing novel stimulation protocols for transcranial stimulations (TMS, tDCS) to innovative AD models exploring the causal relationship between homeostatic imbalance, network decay, and cognitive decline.

Figure 1. Caption: We evaluate our model on n=7 high-resolution structural connectomes of healthy adults. Shown here are the number of connectomes in which US created an engram within the C1/C2 group within the area. Our criterion is that the firing frequency of the readout neuron is larger than three times the standard deviation of its usual firing frequency.
AcknowledgementsThe authors thank the German Federal Ministry of Education and Research and the Hessian Ministry of Science and Research, Art and Culture for supporting this work as part of the NHR funding. Moreover, the authors acknowledge the computing time provided to them on the HPC Lichtenberg II at TU Darmstadt, funded by the German Federal Ministry of Education and Research and the State of Hesse.

References
[1]https://doi.org/10.1016/s0361-9230(99)00182-3
[2] https://doi.org/10.1016/j.neuroimage.2009.10.003
[3]https://doi.org/10.3389/fninf.2024.1323203
[4]https://doi.org/10.1371/journal.pcbi.1003259
[5] https://doi.org/10.1002/hbm.25464

Speakers

Fabian Czappa

Marvin Kaster

Marcus Kaiser

Markus Butz-Ostendorf

Felix Wolf

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P059: Modeling Cholinergic Heterogeneity in Whole-Brain Cortical Dynamics: Bridging Local Circuits to Global State Transitions

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P059 Modeling Cholinergic Heterogeneity in Whole-Brain Cortical Dynamics: Bridging Local Circuits to Global State Transitions

Leonardo Dalla Porta*1, Jan Fousek2, Alain Destexhe3, Maria V. Sanchez-Vives1,4

1Institute of Biomedical Research August Pi i Sunyer (IDIBAPS), Barcelona, Spain
2Central European Institute of Technology (CEITEC), Masaryk University, Brno, Czech Republic
3Institute of Neuroscience (NeuroPSI), Paris-Saclay University, Paris, France
4Catalan Institution for Research and Advanced Studies (ICREA), Barcelona, Spain

*Email: dallaporta@recerca.clinic.cat

Introduction
The wake-sleep cycle consists of fundamentally distinct brain states, including slow-wave sleep (SWS) and wakefulness. During SWS, the cerebral cortex exhibits large, low-frequency fluctuations that propagate as traveling waves [1]. In contrast, wakefulness is characterized by the suppression of low-frequency activity and the emergence of asynchronous, irregular dynamics. While neurotransmitters such as acetylcholine (ACh) are known to regulate the wake-sleep cycle, the mechanisms by which local neuronal interactions give rise to large-scale brain activity patterns are still an open question [2].
Methods
Here, we integrated local circuit properties [2] with global brain dynamics in a whole-brain model [3], constrained by human tractography and cholinergic gene expression. Using a mean-field model, cortical regions incorporated intrinsic excitatory and inhibitory neuronal properties. Connectivity among different brain regions was determined by structural tractography from the human connectome. Cholinergic heterogeneity was introduced using the Allen Human Brain Atlas [4], which quantifies transcriptional activity for over 20,000 genes. M1 and M2 muscarinic receptors, which are targets of ACh, were incorporated by adjusting local node properties, thus creating a detailed virtual brain landscape.
Results
Our model successfully replicated spontaneous slow oscillation patterns and their wave propagation properties, as well as awake-like dynamics. Heterogeneity influenced cortical properties, modulating excitability, synchrony, and the relationship between functional and structural connectivity. Additionally, we quantified global brain complexity in response to stimulation using the Perturbational Complexity Index (PCI) [5] to differentiate brain states and assess the impact of cholinergic heterogeneity on evoked activity. We observed a significant increase in complexity during awake-like states, which depended on the level of heterogeneity.
Discussion
Building on prior insights into cholinergic modulation in local circuits [2], we developed a whole-brain model constrained by muscarinic receptor distributions, bridging intrinsic neuronal properties to large-scale brain activity. Overall, our findings underscore the impact of cholinergic heterogeneity on global brain dynamics and transitions across brain states, shaping the spatiotemporal complexity of neural patterns and functional interactions across cortical areas. Moreover, our approach also offers a pathway to studying the role of various neuromodulators involved in brain state regulation.

Acknowledgements
EU H2020 No. 945539 (Human Brain Project SGA3); INFRASLOW PID2023-152918OB-I00 funded by MICIU / AEI / 10.13039/501100011033/FEDER, UE; ERC SyG grant NEMESIS 101071900
References
1.https://doi.org/10.1523/jneurosci.1318-04.2004
2.https://doi.org/10.1371/journal.pcbi.1011246
3.https://doi.org/10.3389/fncom.2022.1058957
4.https://doi.org/10.1038/nature11405
5.https://doi.org/10.1126/scitranslmed.3006294

Speakers

Maria V. Sanchez-Vives

Leonardo Dalla Porta

Jan Fousek

Alain Destexhe

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P060: Neural Dynamics and Non-Linear Integration: Computational Insights for Next-Generation Visual Prosthetics

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P060 Neural Dynamics and Non-Linear Integration: Computational Insights for Next-Generation Visual Prosthetics

Tanguy Damart*1, Jan Antolík1

1Faculty of Mathematics and Physics, Charles University, Prague, The Czech Republic

*Email: tanguy.damart@protonmail.com
Introduction

Eliciting percepts akin to natural vision using brain computer interfaces is the holy grail of vision prosthetics. However, progress has been slowed by our lack of understanding of how external perturbations, such as electrical stimulation via multi-electrode arrays (MEAs), might perturb the recurrent cortical dynamics and engage the inherent visual representations embedded in the cortical circuitry. Furthermore, investigating these questions directly remains difficult as we rarely have the opportunity to probe the human cortexin-vivo. From this limitation and thanks to the current exponential increase in computing capabilities, modeling and simulation tools naturally came to complement experimental studies.
Methods
We present here a model of intracortical microstimulation (ICMS) applied to a model of columnar primary visual cortex (V1) [1]. The V1 model, built from point neuron models, contains functional retinotopy and orientation maps which are both essential for studying the interaction between external drives such as ICMS and structured spontaneous dynamics. The ICMS is modeled through a phenomenological representation of a MEA that, when activated, causes ectopic spikes in the surrounding cells. The model reproduces two key features of ICMS: sparse and distributed recruitment of neurons, and ectopic spike induction in activated neurons.
Results
We demonstrate that our model reproduces the stereotypical dynamics in V1 seen as a response to ICMS: a transient excitation followed by a lasting inhibition. Comparing the population activity induced by ICMS to the one induced as a response to drifting gratings, we show that ICMS targeting specific orientation columns moderately biases the population activity toward a representation of this orientation. Activating multiple electrodes leads to a slight increase in that orientation bias and produces non-linear activation that could not be predicted by simply adding single-electrode effects. Finally, training a decoder model on responses of the model to natural images, we are also able to show what activity induced by ICMS looks like.
Discussion
Current visual prosthetics rely on phosphene-based encoding through intracortical microstimulation, but this approach underutilizes the complex dynamics of the visual cortex. By investigating how ICMS-induced activity in V1 relates to natural visual activity, we show that current ICMS methods are unlikely to produce anything other than phosphenes and that the non-linear spatio-temporal integrative properties of V1 could be leveraged to enhance visual prosthetic outcomes beyond the resolution limitations of current multi-electrode arrays. The computational framework we developed also enables systematic exploration of stimulation parameters without invasive procedures, such as the development of closed-loop stimulation protocols.

Acknowledgements
The publication was supported by ERDF-Project Brain dynamics, No. CZ.02.01.01/00/22_008/0004643.
References
1. https://doi.org/10.1371/journal.pcbi.1012342

Speakers

Tanguy Damart

Jan Antolik

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P061: A Multi-Scale Virtual Mouse Brain for Investigating Cerebellar-Related Ataxic Alterations

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P061 A Multi-Scale Virtual Mouse Brain for Investigating Cerebellar-Related Ataxic Alterations

Marialaura De Grazia1∗, Elen Bergamo1, Dimitri Rodarie1, Alberto A. Vergani1, Egidio D’Angelo1,2, Claudia Casellato1

1Department of Brain and Behavioural Sciences, University of Pavia, Pavia, Italy
2Digital Neuroscience Center IRCCS Mondino Foundation, Pavia, Italy
∗Email:marialaura.degrazia01@universitadipavia.it

Introduction
Ataxias are neurodegenerative disorders commonly associated with cerebellar dysfunction, often resulting from impaired Purkinje cell (PC) function and progressive loss. In this project, we employed a spiking neural network (SNN) of a mouse olivocerebellar microcircuit, in which we incorporated key PC ataxia- related alterations. We investigated the effect of reduced dendritic arborization, cell shrinkage, and cell loss. These modifications lead to abnormal dynamics within the deep cerebellar nuclei (DCN), which project to the cerebral cortex. Our aim is to create a multiscale framework that integrates cerebellar SNNs with a virtual mouse brain model to investigate the effects of ataxic alterations on whole-brain dynamics (Fig.1A).

Methods
We built a virtual mouse brain network [2] using Allen Mouse Connectome [3] to link neural mass models (a Wong-Wang two-population model per node) on The Virtual Brain (TVB) platform (Fig.1B). Network parameters were tuned employing resting-state fMRI of 20 mice [4]. We are testing a TVB co-simulation framework [5] by replacing each cerebellar node with a cerebellar SNN. For cerebellar network reconstruction and simulation, we used the Brain Scaffold Builder (BSB) [1], which integrates the NEST simulator (Fig.1C). After validating healthy dynamics, we introduced ataxia-related alterations (reduced PC dendritic complexity, shrinkage, and density) and tested various stimulation protocols (e.g. Poisson inputs to mossy fibers from 4 to 100 Hz).

Results
The results indicate that as PC density, dendritic complexity index (DCI) and size decrease, the DCN become increasingly disinhibited due to reduced inhibitory input from PCs. The mildest network dysfunction occurs with DCI reduction alone, while more pronounced changes emerge when PCs also shrink. However, the most substantial disruptions in cerebellar dynamics arise with progressive PC density reduction (Fig.1D).Additionally, TVB global coupling and Wong-Wang model parameters were optimized for each resting-state network to maximize the match between experimental and simulated functional connectivity matrices. TVB simulations are in progress.

Discussion
Next steps will consist in further investigation of the dynamics of the ataxic cerebellar SNN, with a particular focus on exploring the electrophysiological changes within the PC model.Moreover, we are testing a TVB-NEST co-simulation framework and tuning the proxy nodes, the interface nodes between the two simulators, that enable the bidirectional conversion between spike-based and rate-coded information. This multiscale model will enhance our ability to predict and analyze alterations in large-scale brain activity and functional networks under ataxic conditions. Furthermore, it may serve as a computational tool for evaluating neuromodulation protocols (e.g. Transcranial Magnetic Stimulation) for treating cerebellar ataxias.

Figure 1. Figure 1: A. Multiscale framework for cerebellar SNN-neural mass interaction. B. TVB integrates Mouse Connectome with Wong-Wang models. C. Cerebellar network built by BSB maps SNN placement and connectivity. D. Simulating ataxia: reduced PC DCI affects granule cells to PC (via pf: parallel fibers) connectivity, PC loss impacts PC-DCNp connectivity. Testing: mossy fibers input vs. DCNp firing rate.
Acknowledgements

PRIN project 20228B2HN5 “cerebellar NEuromodulation in ATaxia: digital cerebellar twin to predict the MOVEment rescue (NEAT-MOVE)” (CUP master: F53D23005950006310)
References

1.https://doi.org/10.1038/s42003-022-04213-y
2.https://doi.org/10.1523/ENEURO.0111-17.2017

3. doi:10.1038/nature13186
4.https://doi.org/10.1038/s41467-021-26131-z
5.https://doi.org/10.1016/j.neuroimage.2022.118973

Speakers

Claudia Casellato

Alberto Arturo Vergani

Research Fellow, University of Pavia

Egidio D'Angelo

Marialaura De Grazia

Dimitri Rodarie

Elen Bergamo

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P062: A complete computational model to explore human sound localization

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P062 A complete computational model to explore human sound localization

Francesco De Santis1, Paolo Marzolo1,AlessandraPedrocchi1,AlbertoAntonietti1
1 Department of Electronics, InformationandBioengineering,Politecnicodi Milano, Milano, Italy
* Email:francesco.desantis@polimi.it
Introduction

Animal ability to localize sounds in space is one of the most studied aspects of hearing.Sound source position is derived from interaural time difference (ITD), interaural level difference (ILD), and spectral cues.Despite decades of auditory neuroscience research, critical questionsremainabout the neural processes supportinghuman sound localization. Its understanding isparticularlyacute for cochlear implant users, whose devices oftenfail toprovide precise spatialperception.Our aimis to address these interrogatives through the implementation of a comprehensive spiking neural network.

Methods
Themodel(depicted in Fig. 1)is composed of a peripheral section, from the sound to the spiking output of the cochlea, and a neuralsection,from the auditory nerve fibers to the superior olivary complexnuclei,developedusingBrian2Hears[1]andNEST[2]neuralsimulatorrespectively.The main inputs to the network are sounds used inin-vivoexperiments in mammals, such as pure tones at different frequencies, clicks, and white noises.To evaluate how source positionimpactedthe overall model activity, we provided stimuli of 1 s duration from different spatial positions in the frontal azimuth plane,analyzing the corresponding spike distribution and overall firing rate of all thein-silicopopulations involved.Special attention was given to the activity of the lateral and medial superior olives(LSO and MSO), two nuclei of the superior olivary complex considered to be the main players in the processing of ILDsand ITDs.
Results
The wide range of our model offered the possibility of facing various validation sites, comparing in-silico activity with different results obtained experimentallyin-vivoorin-vitro. First, allneuralpopulations showed a phase-locked spikingactivity,witha refinementforhigher-level populationsfundamental for correct ITD processing[3]. The analysis of the overall population firing rate of LSO and MSO also showed physiological plausibility, with respectively an ipsilateral-increasing and a contralateral-increasing sigmoid-like behavior in response to shifting azimuth location[3,4]. Finally, the reproduction of specific experimental setups focused on the MSO processing of ITDs showed coherent results in the effect of inhibitory input blockage[5]and in input delay manipulation on the overall MSO activity[6].
Discussion
Theimplementedcomputationalmodeladdressessome of the theoriesconcerningthe processing of sound and thecomputationofitslocationatthebrainstemlevelinhumans.Webelievethatourmodelcouldbe apromisingvalidationbase forstudyingtheeffectofcochlearimplant-generatedartificialinputs for soundlocalization,sheddinglight on thedifferentresponseof theinvolvedauditoryneuronswithrespectto arealsoundstimulation.

Figure 1. End-to-end spiking neural network
Acknowledgements
The work of AA, AP, and FDS in this research is supported by EBRAINS-Italy (European BrainReseArchINfrastructureS-Italy), granted by the Italian National Recovery and Resilience Plan (NRRP), M4C2, funded by the European Union –NextGenerationEU(Project IR0000011, CUP B51E22000150006, EBRAINS-Italy).
References
●
https://doi.org/10.3389/fninf.2011.00009

●
https://doi.org/10.4249/scholarpedia.1430

●
https://doi.org/10.1002/cphy.c180036

●
https://doi.org/10.1152/physrev.00026.2009

●
https://doi.org/10.1038/ncomms4790

●
https://doi.org/10.1523/JNEUROSCI.1660-08.2008

Speakers

Alberto Antonietti

Francesco De Santis

Paolo Marzolo

Alessandra Pedrocchi

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P063: Dichotomous Dynamics of Hippocampal and Lateral Septum Oscillations: State-Dependent Topology and Directional Causality During Sleep and Wakefulness

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P063 Dichotomous Dynamics of Hippocampal and Lateral Septum Oscillations: State-Dependent Topology and Directional Causality During Sleep and Wakefulness

Amir Khani1,Nima Dehghani1,2*

1 N3HUB Initiative, Massachusetts Institute for Technology, Cambridge, U.S.A.
2McGovern Institute for Brain Research, Massachusetts Institute for Technology, Cambridge, U.S.A.

*Email: nima.dehghani@mit.edu

Introduction

Sharp-wave ripples (SWRs) in the hippocampus (HPC) and high-frequency oscillations (HFOs) in the lateral septum (LS) play critical roles in memory consolidation and information routing to subcortical areas. However, the precise spatiotemporal dynamics and causal relationships between these oscillations remain poorly understood. Using multiple analytical approaches, we explored the coordination of HPC-SWR and LS-HFO oscillations during Non-Rapid Eye Movement (NREM) sleep and wakefulness, focusing on their topological features, causal relationships, and dimensional properties.
Methods
We analyzed publicly available LFP recordings from hippocampal subfields and the lateral septum in freely behaving rats [1]. To identify oscillations, we detected ripples following the methods described in [1]. To assess temporal coordination, we employed conditional probability analysis to quantify ripple co-occurrence between regions. To characterize oscillation structure, we applied Topological Data Analysis (TDA) using time-delay embedding (dimension = 3, delay = 2). To determine directional influences, we implemented Convergent Cross Mapping (CCM) for causality assessment [2]. To evaluate the dimensionality of neural activity, we utilized Principal Component Analysis (PCA) across individual channels, regions, and brain states [3].
Results
HPC ripples consistently preceded LS ripples, with the conditional probability of LS ripples given HPC-SWR, P(LS|HPC), higher than the probability of HPC-SWR given LS ripples, P(HPC|LS), especially during NREM sleep (Fig.1E). TDA revealed distinct topological structures: LS HFOs showed state-dependent complexity differences between sleep and awake, while HPC ripples maintained similar features across states (Fig.1D). Bidirectional causality analysis showed LS-HFOs influenced HPC-SWRs more than the reverse across both states, with a stronger relationship during NREM sleep (Fig.1C). Dimensionality analysis, examining SWR events across epochs/channels and applying PCA, highlighted the variability and complexity of SWRs in HPC compared to more uniform LS HFOs (Fig.1A,F).
Discussion
Our findings reveal a complex, bidirectional relationship between HPC and LS during ripple events, with stronger coupling during NREM sleep. The higher intrinsic dimensionality of HPC activity during SWRs reflects its role in complex memory processes, while the lower-dimensional LS activity suggests a streamlined relay function [1]. These results align with prior evidence showing LS neuron activation by hippocampal SWRs [1] and highlight state-dependent coordination between HPC and LS. State-dependent coordination changes suggest that during NREM sleep, the coordination supports memory consolidation, while during wakefulness, it facilitates spatial navigation and behavior.

Figure 1. (A) PCA dimensionality of HPC/LS ripples during NREM sleep (left) and wakefulness (right). (B) Raw LFP traces of HPC-SWR (top) and LS-HFO (bottom). (C) Bidirectional CCM analysis: NREM (top) and wakefulness (bottom). (D) Topological features during NREM: H1 count (left) and Shannon entropy (right). (E) Ripple co-occurrence probability: NREM (left) and wakefulness (right). (F) Channel-wise PCA dime
Acknowledgements
N.D. is supported by NIH Grant R24MH117295.The authors wish to thank NIH for its sponsorship of DANDI archive (DANDI: Distributed Archives for Neurophysiology Data Integration), which provided the open-access data used in this study.
References
[1] Tingley, D., Buzsaki, G. (2020). Routing of hippocampal ripples to subcortical structures via the
lateral septum. Neuron, 105(1), 138-149.e5.
[2] Sugihara, G., et al (2012). Detecting
causality in complex ecosystems. Science, 338(6106), 496–500.
[3] Dehghani, N., et al (2010). Magnetoencephalography
demonstrates multiple asynchronous generators during human sleep spindles. Journal of
Neurophysiology, 104(1), 179-188.
[4] Tingley, D., Buzsaki, G. (2018). Transformation of a spatial map across the hippocampal-lateral septal
circuit. Neuron, 98(6), 1229-1242.e5.

Speakers

Nima Dehghani

MIT

Amir Khani

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P064: Anatomically and Functionally Constrained Bio-Inspired Recurrent Neural Networks Outperform Traditional RNN Models

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P064 Anatomically and Functionally Constrained Bio-Inspired Recurrent Neural Networks Outperform Traditional RNN Models

Mo Shakiba1,2, Rana Rokni1,2, Mohammad Mohammadi1,2,Nima Dehghani2,3*

1Neuromatch Academy, Neuromatch, Inc.
2N3HUB Initiative, Massachusetts Institute for Technology, Cambridge, U.S.A.
3McGovern Institute for Brain Research, Massachusetts Institute for Technology, Cambridge, U.S.A.

*Email: nima.dehghani@mit.edu

Introduction
Understanding how neural circuits drive sensory processing and decision-making is a central neuroscience challenge. Traditional Recurrent Neural Networks (RNNs) capture temporal dynamics but fail to represent the structured synaptic architecture seen in biological systems [1]. Recent spatially embedded RNNs (seRNNs) add spatial constraints for better biological relevance [2], yet they do not fully exploit detailed anatomical and functional data to enhance task performance and neural alignment.
Methods
We introduce a bio-inspired RNN that integrates detailed anatomical connectivity and two-photon calcium imaging data from the MICrONS dataset (https://www.microns-explorer.org/cortical-mm3), which offers nanometer-scale reconstructions and functional recordings from mouse visual cortex. Using neuronal positions, synaptic connections, functional correlations, and Spike Time Tiling Coefficients (STTC) [3]—a robust metric that eliminates firing rate biases—we constrain our model with biologically informed weight initialization, communicability calculations, and a regularizer that penalizes long-distance connections while boosting communicability to promote realistic network properties.
Results

Trained on three distinct decision-making tasks—a 1-step inference task, a Go/No-Go task, and a perceptual decision-making task— our bio-inspired RNN demonstrated significant performance improvements over baseline models across 30 simulations per model (900 total simulations across all model variants). Variants combined W* (biologically initialized weights) or W (standard initialization), D* (actual neuron distances) or D (random distances), and C (communicability calculation). Specifically, the anatomically and functionally constrained model (W*D*C) achieved the highest average accuracy across all tasks: 89.4% on the 1-step inference task, 96.9% on the Go/No-Go task, and 86.7% on the perceptual decision-making task.
Moreover, the biologically constrained model demonstrated superior performance across other evaluation metrics, including validation accuracy, training and validation loss, and network properties such as modularity and small-worldness. Specifically, the average modularity of the W*D*C and WD*C models was highest across all tasks, with values of 0.583 (1-Step Inference), 0.558 (Go/NoGo), and 0.594 (Perceptual Decision Making). Similarly, the average small-worldness was also the highest across two tasks, with values of 3.513 (Go/NoGo), and 4.325 (Perceptual Decision Making) (Fig. 1c-e)
Discussion

Our findings demonstrate that incorporating biological constraints into RNNs significantly boosts both task performance and the emergence of realistic network properties, mirroring actual neural architectures. Future work should extend this approach to visual processing tasks, explore other architectures such as LSTMs and GNNs, and integrate additional biological constraints.

Figure 1. (a) Weight initialization matrix (top left) from MICrONS data, combining functional correlation (bottom left) and STTC (bottom right) with log-normal noise. Anatomical distance matrix (top right) shows neuron positioning. (b) Top 10% models (900 simulations): Effects of λ and W on accuracy, loss, modularity, and small-worldness. (c-e) Model variants task performance shows WDC outperforming RNNs.
Acknowledgements
N.D. is supported by NIH Grant R24MH117295. The authors thank NIH for sponsoring DANDI archive, which provided the open-access data used in this study. M.S., R.R., and M.M. thank Neuromatch Academy for its support and resources for young scholars and this study. They also thank the DataJoint team for their help and guidance.
References
● Perich, M. G., & Rajan, K. (2020). Rethinking brain-wide interactions through multi-region ‘network of networks’ models. Current Opinion in Neurobiology, 65, 146–151.
● Achterberg, J., et al. (2023). Spatially embedded recurrent neural networks reveal widespread links between structural and functional neuroscience findings. Nature Machine Intelligence, 5(12), 1369–1381.
● Cutts, C. S., & Eglen, S. J. (2014). Detecting Pairwise Correlations in Spike Trains: An Objective Comparison of Methods and Application to the Study of Retinal Waves. The Journal of Neuroscience, 34(43), 14288–14303.

Speakers

Nima Dehghani

MIT

Mo Shakiba

Rana Rokni

Mohammad Mohammadi

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P065: Jaxley: Differentiable simulation enables large-scale training of detailed biophysical models of neural dynamics

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P065 Jaxley: Differentiable simulation enables large-scale training of detailed biophysical models of neural dynamics

Michael Deistler*1,2, Kyra L. Kadhim2,3, Matthijs Pals1,2, Jonas Beck2,3, Ziwei Huang2,3, Manuel Gloeckler1,2, Janne K. Lappalainen1,2, Cornelius Schröder1,2, Philipp Berens2,3, Pedro J. Gonçalves1,2,4,5, Jakob H. Macke*1,2,6

1Machine Learning in Science, University of Tübingen, Germany
2Tübingen AI Center, Tübingen, Germany
3Hertie Institute for AI in Brain Health, University of Tübingen, Tübingen, Germany
4VIB-Neuroelectronics Research Flanders (NERF)
5imec, Belgium
6Department Empirical Inference, Max Planck Institute for Intelligent Systems, Tübingen, Germany

*Email:michael.deistler@uni-tuebingen.de, jakob.macke@uni-tuebingen.de
Introduction

Biophysical neuron models provide mechanistic insight underlying empirically observed phenomena. However, optimizing the parameters of biophysical simulations is notoriously difficult, preventing the fitting of these models to physiologically meaningful tasks or datasets. Indeed, current fitting methods for biophysical models are typically limited to a few dozen parameters [1]. At the same time, backpropagation of error (backprop) has enabled deep neural networks to scale to millions of parameters and large datasets. Unfortunately, no current toolbox for biophysical simulation can perform backprop [2], limiting any study of whether backprop could also be used to construct and train large-scale biophysical neuron models.

Methods
We built a new simulation toolbox, Jaxley, which overcomes previous limitations in constructing and fitting biophysical models. Jaxley implements numerical solvers required for biophysical simulations in the machine learning library JAX. Thanks to this, Jaxley can simulate biophysical neuron models and it can compute the gradient of such simulations with backpropagation of error (Fig. 1a). This makes it possible to optimize thousands of parameters of biophysical models with gradient descent. In addition, Jaxley can parallelize simulations on GPUs, which speeds up simulation by at least two orders of magnitude (Fig. 1b).

Results
We applied Jaxley to a range of datasets and models. First, we applied Jaxley to a series of single neuron tasks and found that it outperforms gradient-free optimization methods (Fig. 1c). Next, we built a simplified biophysical model of the retina (Fig. 1d). We optimized synaptic and channel conductances on dendritic calcium recordings and found that the trained model exhibits compartmentalized responses (matching experimental recordings [3]). Third, we built a recurrent neural network model with biophysically-detailed neurons and trained this network on working memory tasks. Finally, we trained a network of morphologically detailed neurons to solve MNIST with 100k biophysical parameters (Fig. 1e).

Discussion
Optimizing parameters of biophysically detailed models is challenging, and previous (gradient-free) methods have been limited to a few dozen parameters. We developed Jaxley, which overcomes these limitations. Jaxley implements numerical solvers required for biophysical simulations [4], it can easily parallelize simulations on GPUs, and it can perform backprop. Together, these features make it possible to construct and optimize large neural systems with thousands of parameters. We designed Jaxley to be easy to use and we provide extensive documentation, which will make it easy for the community to adopt the toolbox. Jaxley bridges systems neuroscience and biophysics and will enable new insights and opportunities for multiscale neuroscience.

Figure 1. (a) Jaxley can compute gradients with backprop. (b) Jaxley is as accurate as the NEURON simulator and can achieve speed-ups with GPU parallelization. (c) Jaxley can identify single-neuron models, sometimes much more efficient than a genetic algorithm. (d) Biophysical model of mouse retina predicts dendritic calcium response. (e) Biophysical network solves MNIST computer vision task.
Acknowledgements
This work was supported by the German Research Foundation (DFG) through Germany’s Excellence Strategy (EXC 2064 – PN 390727645) and the CRC 1233 "Robust Vision", the German Federal Ministry of Education and Research (FKZ: 01IS18039A), the 'Certification and Foundations of Safe Machine Learning Systems in Healthcare' project, and the European Union (ref. 101089288, ref. 101039115).

References
[1] Van Geit, W., De Schutter, E., & Achard, P. (2008). Automated neuron model optimization techniques: a review.Biological cybernetics,99, 241-251.
[2]Hines, M. L., & Carnevale, N. T. (1997). The NEURON simulation environment.Neural computation,9(6), 1179-1209
[3]Ran, Y., Huang, Z., Baden, T., Schubert, T., Baayen, H., Berens, P., ... & Euler, T. (2020). Type-specific dendritic integration in mouse retinal ganglion cells.Nature communications,11(1), 2101.
[4]Hines, M. (1984). Efficient computation of branched nerve equations.International journal of bio-medical computing,15(1), 69-76.

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P066: In-silico study on the dynamical and topological relationships between three-dimensional cultures and relative slices

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P066 In-silico study on the dynamical and topological relationships between three-dimensional cultures and relative slices

Leonardo Della Mea1*, Angelo Piga2,3, Jordi Soriano3

1DIBRIS, University of Genoa, Genoa, Italy
2Department of Economics and Managements, University of Pisa, Pisa, Italy
3Institute of Complex Systems, University of Barcelona, Barcelona, Spain

*Email: leonardo.dellamea@edu.unige.it

Introduction

The in-vitro three-dimensional neuronal culture represents a pioneering technological advancement in exploring brain function and dysfunction in a more realistic environment[1], [2]. The recording of the entire network activity remains a challenge, since researches resort to methods developed for two-dimensional cultures, such as multi-electrode arrays or calcium fluorescence imaging. The question of whether the read-out layer, through which the network is recorded, reliably captures topological and dynamical properties, quickly arises. In the this study we utilized in-silico modelling of developing 3D neuronal networks to assess the reliability of a single layer of the culture in capturing dynamical and topological features of the entire parent network.

Methods
The networks were constructed through the random placement of neurons within a rectangular prism. The cell’s dendritic and axonal domains were constructed upon a main trunk, consisting in concatenated segments, and a group of arborizations, depicted by spherical regions. The overlap of different cells’ dendritic and axonal arbours results in synaptogenesis. The network is then simulated as a pulsed coupled neuronal network embedding the Izhikevich model[3]. The bottom layer of neurons were drawn out to emulate the MEA recordings. Its dynamical properties -focused on the features of the network burst (NB)– and topological traits –based on small-worldness (SW)[4]and modularity (Q)[5]- were compared to the one of the entire parent cultures.

Results and discussion
From a dynamical and topological perspective, statistically significant differences were observed for all the parameters measured.For the network’s slice, the mean NB sizes and dynamical variability are regularly overestimated, whereas the NB duration is underestimated. Due to slicing, a variable fraction of the neurons in the layer are exposed to the propagating front of the burst, thus, the dynamical differences observed in slices may be due the fact that NB events are very unlikely to systematically engage equal fractions of the sub-network, justifying the higher dynamical variability. In addition, the reduced size of the network makes the slices liable to wrongly capture the mean event sizes and duration; indeed, both measures depend on the network size. Modularity exhibited a monotonic decline in both 3D and slice systems, although it was marginally overestimated in the slice. The 3D network shows a bell-shaped trendof SW valuesacross the maturation, peaking in the middle of developmental phase. In contrast, the slice’s values differed, consistently under-estimating it.Sampling the edges from a network whose architecture is grounded on distance-dependent probability of connection, results in sub-networks where this feature is exacerbated. Consequently, in slices, communities are more starkly outlined and in turn Q increases and SW decreases -due to the reduction of shortcuts.

Acknowledgements
The author wish to thank Prof. Jordi Soriano and Angelo Piga , for their kind advice on the experimental procedure and useful discussion. The author declare no use of Artificial Intelligence in this study.
References
[1] https://doi: 10.1016/j.isci.2020.101434.
[2] https://doi: 10.1002/term.2508.
[3] https://doi: 10.1109/TNN.2003.820440.
[4] https://doi: 10.1016/j.neuroimage.2009.10.003.

[5]https://doi: 10.1103/PhysRevE.70.066111.

Speakers

Leonardo Della Mea

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P067: Optimization-Based Insights into Network Configurations of the Songbird Premotor Cortex

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P067 Optimization-Based Insights into Network Configurations of the Songbird Premotor Cortex

Fatima M. Dia*1, Maher Nouiehed2, Arij Daou1

1Department of Biomedical Engineering, American University of Beirut, Beirut, Lebanon
2Department of Industrial Engineering and Management, American University of Beirut, Beirut, Lebanon

*Email:fmd14@mail.aub.edu

Introduction
Neural circuits in the brain maintain a delicate equilibrium between excitation and inhibition, yet how this balance operates remains unclear[1]. Moreover, neural circuits often exhibit sequences of activity that rely on excitation and inhibition, but the contribution of local networks to their generation is not well understood. This study investigates neural sequence generation within the High Vocal Center (HVC) of the adult zebra finch forebrain. The HVC plays a critical role in the execution of temporally precise courtship songs and is comprised of three neural populations with distinct electrophysiological responses: glutamatergic basal ganglia–projecting (HVCX) and forebrain-projecting (HVCRA) cortical neurons, and GABAergic interneurons (HVCINT)[2]. While the connections between these neuronal classes are known[1,3], how they orchestrate this temporally precise neural sequence remains largely unknown.
Methods
To address this question, we applied optimization techniques and mathematical modeling to describe the relationships among HVCRA, HVCX, and HVCINT neurons and their bursting patterns. Our approach focused on uncovering the underlying cytoarchitecture of the HVC neural network by utilizing biologically realistic constraints. These constraints included the pharmacological nature of synaptic connections, anatomical and intrinsic properties, neuronal population ratios, precise burst timing, and spiking frequency during song motifs[2,4]. The study incorporated both closed and open network configurations to assess their ability to reproduce observed bursting sequences.
Results
Our computational framework successfully predicted the minimalistic synaptic connections required to replicate the observed bursting patterns of the HVC network. The model identified specific network topologies that satisfied experimental constraints while maintaining functional output. Additionally, our findings indicated that certain network configurations necessitate additional nodes to form a fully connected network capable of sustaining stable sequential bursting. These predictions align with previous experimental data and provide novel insights into potential connectivity motifs that could underlie the temporal precision of song production.
Discussion
This study bridges experimental data with computational predictions, offering a framework for understanding how local excitatory and inhibitory interactions within HVC generate precise neural sequences. By identifying minimal network configurations, our model provides a hypothesis regarding the synaptic architecture required for sequence generation. Future work should incorporate in vivo validation of the predicted connectivity patterns using electrophysiological and optogenetic approaches. Our findings contribute to a broader understanding of how premotor circuits coordinate motor behaviors and may have implications for studying sequence generation in other brain regions beyond the songbird HVC.

Acknowledgements
This work was supported by the University Research Board (URB) and the Medical Practice Plan (MPP) grants at the American University of Beirut.
References
● https://doi.org/10.1152/jn.00162.2013
● https://doi.org/10.1038/nature00974
● https://doi.org/10.1038/nature09514
● https://doi.org/10.1152/jn.00952.2006

Speakers

Fatima Dia

Maher Nouiehed

Arij Daou

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P068: Modelling the dynamics of in vitro cultured neural networks by using biophysical neuronal models

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P068 Modelling the dynamics of in vitro cultured neural networks by using biophysical neuronal models

Marco Fabiani1, Ludovico Iannello3, Fabrizio Tonelli4, Eleonora Crocco4,
Federico Cremisi4, Lucio Calcagnile2, Riccardo Mannella1, Angelo Di Garbo1,2
1Department of Physics, University of Pisa
2Institute of Biophysics - IBF, CNR of Pisa
3Institute of Information Science and Technologies ”Alessandro Faedo” - ISTI,
CNR of Pisa
4Scuola Normale Superiore - SNS, Pisa
Email: m.fabiani5@studenti.unipi.it, angelo.digarbo@ibf.cnr.it
IntroductionIn this contribution we study the dynamical behaviours arising in a biophys-ical inspired neuronal network of excitatory and inhibitory neurons. The setup of the corresponding model was done by using the electro-physiological data recorded on cultured neuronal networks. The recordingsof the local field potential generated by the neurons was carried outby using multielectrode array (MEA) apparatus [2]. In particular, weinvestigated the dynamics emerging in a cultured population of (Mapk/erkinhibition and BMP inhibition, MiBi) neurons of the entorhinal cortex [1].
MethodsThe MEA recordings were obtained from a grid of 64x 64 electrodes coveringan area of 3.8 mm x 3.8 mm of the neuronal culture. The correspondinglocal field potentials were acquired with a sampling frequency of 20 kHz.The spiking times of the cultured neurons were obtained by applying specificalgorithms to the local field potential signals. Then, an artificialbiophysical inspired neural network was built by employing Hodgkin-Huxley-type models for the single neuron.Finally,the parameters describing the computational neural network were chosen byrequiring that the simulation results were qualitatively in agreement with thecorresponding experimental data.
ResultsAccording to the results described in [2, 3] we found that the MiBi culturedneuoronal network is capable of generating bursting activity. Moreoverthe analysis of these data show that the bursting activity is triggered bysome points of the cultured network (center of activity). In addition, thepropagation on the neural culture was characterized by the center of activitytrajectories (CAT). Furthermore, these cultures exhibit neuronalavalanches with power decay. We have shown that the computationalmodel is capable of reproducing the bursting dynamics observed in vivo cul-tured neural network by choosing suitable parameter values in an all-to-allcoupled network.By setting up a more detailed network model, obtained by modifying theconnectivity matrix and the density of neurons, we proved that such a neu-ronal network is capable of reproducing many of the experimental data and,qualitatively, their specific features.
DiscussionAlthough the mathematical model has some intrinsic limitations, the corre-sponding numerical results helped us to shed light on some basic mechanismsresponsible for the generation of bursting in the network and this could beused to infer that such processes should be present also in the MiBi culturedneuronal network. It would be interesting to check if improving the qualityof the neuronal model will be sufficient to reproduce others experimental fea-tures that are not captured by the adopted model. This include, for instance,to use more realistic single neuron model, synaptic connectivity and synapticplasticity.

Acknowledgements
The research was in part supported by the Matteo Caleo Foundation, by Scuola Normale
Superiore (FC), by the PRIN AICult grant #2022M95RC7 from the Italian Ministry of
University and Research (MUR) (FC) and by the Tuscany Health Ecosystem - THE grant
from MUR (FC, GA, ADG).
References
[1]Tonelli F. et al. “Dual inhibition of MAPK/ERK and BMP signaling
induces entorhinal-like identity in mouse ESC-derived pallial progeni-
tors.” In: Stem Cell (2025). doi: 10.1016/j.stemcr.2024.12.002.
[2] Ludovico Iannello et al. “Analysis of MEA recordings in cultured neural
networks”. In: (2024), pp. 1–5. doi: 10.1109/COMPENG60905.2024.10741515.
[3] Ludovico Iannello et al. “Criticality in neural cultures: Insights into
memory and connectivity in entorhinal-hippocampal networks”. In:
Chaos, Solitons and Fractals 194 (2025), p. 116184.

Speakers

Angelo Digarbo

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P069: Statistics of spiking neural networks based on counting processes

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P069 Statistics of spiking neural networks based on counting processes

Anne-Marie Greggs1*, Alexander Dimitrov1

1Department of Mathematics and Statistics, Washington State University, Vancouver, WA

*Email: anne-marie.greggs@wsu.edu
Introduction

Estimating neuronal network activity as point processes is challenging due to the singular nature of events and high signal dimensionality[1]. This project analyzes spiking neural networks (SNNs) using counting process statistics, which are equivalent integral representations of point processes[2]. A small SNN of Leaky Integrate-and-Fire (LIF) neurons is simulated, and spiking events are counted as a vector counting process N(t). The Poisson counting process has known dynamic statistics over time: both mean(t) and variance(t) are proportional to time (= r_i*t for each independent source with rate r_i). By standardizing the data, mean dynamics and heteroscedasticity can be removed, allowing comparison to a baseline Poisson counting process.
Methods
Using Brian2[3], an SNN with LIF neurons and Poisson inputs is simulated. Independent and correlated Poisson processes are modeled, generating spike trains for analysis. The counting process, a stochastic process producing the number of events within a time period, is analyzed using the vector counting process. Mean and covariance of spiking events are estimated for both SNN and Poisson processes, facilitating comparison of statistical properties after standardization by subtracting the mean and scaling by the standard deviation to account for temporal dependencies.
Results
Fig 1 shows the simulated spiking dynamics of two neurons over time. The standardized counts indicate variability aligned with Poisson statistical properties. While mean counts show a consistent trend, variance reflects the stochastic nature of neural activity. The centered plot's standard deviation equals the square root of rate and time. The standardized plot’s standard deviation equals 1, serving as a comparison template, starting at 200 milliseconds to avoid biases when rescaling initial small counts.
The covariance matrix quantifies relationships between neurons at certain time and activity levels. Comparing the SNN to modeled Poisson processes reveals notable differences in covariance structures, with the SNN demonstrating greater inter-unit correlation.

Discussion
This study establishes a framework for analyzing the statistical properties of neural network activity, enabling researchers to gain insights into the dynamics of spiking networks. Understanding these aspects is crucial for examining how neural networks respond to stimuli and adapt to changing environments.
The findings highlight the importance of inter-unit dependencies in neural data, with the proposed estimators effectively capturing these dynamics. Future research should broaden parameter exploration and apply the estimators to complex models and real-world data, including comparisons between inhomogeneous Poisson processes with time-varying rates, temporal dependencies, and non-Poisson processes of SNNs.

Figure 1. Two 3D plots compare the activity of Neuron 1 and Neuron 2 over time. The left plot shows counts centered based on the theoretical expectations, while the right plot shows counts standardized by both the theoretical expectations and theoretical standard deviations, with multiple lines representing each of 100 samples.
Acknowledgements
N/A
References
● Brown, E. N., Kass, R. E., & Mitra, P. P. (2004). Multiple neural spike train data analysis: state-of-the-art and future challenges. Nature Neuroscience, 7(5), 456-461. doi: 10.1038/nn1254
● Cox, D. R., & Isham, V. (1980). Point processes. Chapman and Hall.
● Stimberg M, Brette R, Goodman DFM (2019). Brian 2, an intuitive and efficient neural simulator. eLife 8:e47314. doi: 10.7554/eLife.47314

Speakers

Alexander Dimitrov

Anne-Marie Greggs

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P070: Neurospheroids as building blocks for 3D brain-like structures

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P070 Neurospheroids as building blocks for 3D brain-like structures

Ilaria Donati della Lunga*,1, Francesca Callegari1, Fabio Poggio1, Letizia Cerutti1, 2, Mattia Pesce2, Giovanni Lobello1, Alessandro Simi3, Mariateresa Tedesco1, Paolo Massobrio1,4, Martina Brofiga1,2,5

1Department of Informatics, Bioengineering, Robotics and Systems Engineering (DIBRIS), University of Genova, Genova, Italy
2Neurofacility, Istituto Italiano di Tecnologia (IIT), Genova, Italy
3 Central RNA Laboratory, Istituto Italiano di Tecnologia (IIT), Genova, Italy
4National Institute for Nuclear Physics (INFN), Genova, Italy
5ScreenNeuroPharm, Sanremo, Italy

* Email:ilaria.donatidellalunga@edu.unige.it
Introduction.Conventionalin vitroneuronal networkshave provided insights into brain function and disease mechanisms [1] but often neglect importantin vivoproperties such as three-dimensionality (3D) and heterogeneity, i.e. the coexistence of different neuronal types. To address these limitations, we aimed to develop 3D cortical (C) and hippocampal (H) cell aggregates known as neurospheroids (NSs): their coupling allowed us to generate homogeneous (CC, HH) or heterogeneous (CH) assembloids (ASs). This study aims to prove that these models enhance the reproducibility, viability, and biological significance of neuronal cultures, while exhibitingin vivo-like electrophysiological patterns, characterized by brain waves.

Methods.We employed a multi-modal approach: structural and mechanical properties were assessed via immunostaining and atomic force microscopy, functional activity was evaluated by calcium imaging and electrophysiological recordings with Micro-Electrode Arrays. To detect neuronal activity and synchronization, fluorescence traces were analyzed using Schmitt trigger and SPIKE-synchronization algorithms. From the electrophysiological activity, we identified the typical observedin vivobrain waves. Spectral analysis was performed using the wavelet transform to assess oscillatory patterns, while the functional E/I ratio metrics [2] was used to validate the physiological relevance of the models.

Results.Morphological analysis revealed a faster geometric expansion and higher cell proliferation in H than C. Stiffness values matchedin vivoconditions [3], and immunostaining confirmed physiological composition and organization [4]. We developed homogeneous (CC, HH) and heterogeneous (CH) ASs coupling pairs of NSs, ensuring physical connections while preserving structural segregation.Calcium dynamic revealed functional intra- and inter-module communication in ASs. Moreover, spectral analysis showed the generation of typical brain waves in our 3D models,with CH displaying different dynamics at DIV 18, marking a transition phase. The excitation/inhibition ratio matched physiological conditions [2].

Discussion.Our findings showed that the developed NSs and ASs enhance physiological relevance by replicating key aspects of brain organization and function. The integration of cortical and hippocampal regions within ASs enables the study of modular and heterogeneous network dynamics. Functional analyses confirm the emergence of complex oscillatory patterns, reflectingin vivo-like network behavior. The ability of ASs to maintain structural segregation while ensuring functional connection makes them a valuable tool for investigating fundamental neurobiological mechanisms, modelling neurodegenerative diseases and testing therapeutic interventions.

Acknowledgements
This work was supported by #NEXTGENERATIONEU (NGEU) and funded by the Ministry of University and Research (MUR), National Recovery and Resilience Plan (NRRP), project MNESYS(PE0000006) - A Multiscale integrated approach to the study of nervous system in health and disease (DN. 1553 11.10.2022).
References
1.https://doi.org/10.1152/jn.00575.2016
2.https://doi.org/10.1038/s41598-020-65500-4
3.https://doi.org/10.1007/s11831-019-09352-w

4.https://doi.org/10.1016/j.cmet.2011.08.016

Speakers

Paolo Massobrio

Associate Professor, Univeristy of Genova

Francesca Callegari

Ilaria Donati della Lunga

Fabio Poggio

Letizia Cerutti

Mattia Pesce

Giovanni Lobello

Alessandro Simi

Mariateresa Tedesco

Martina Brofiga

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P071: The Role of Descending Inferior Colliculus Projections to the Cochlear Nucleus in the Hyperactivity Underlying Tinnitus

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P071 The Role of Descending Inferior Colliculus Projections to the Cochlear Nucleus in the Hyperactivity Underlying Tinnitus

Katherine Doxey*1, Timothy Balmer2, Sharon Crook1

1School of Mathematical and Statistical Sciences, Arizona State University, Tempe, United States
2School of Life Sciences, Arizona State University, Tempe, United States

*Email: kedoxey@asu.edu

Introduction

Tinnitus is the perception of a sound without the presence of auditory stimuli. Over 2.5 million veterans are currently receiving disability benefits for a tinnitus diagnosis [1] and the likelihood of veterans screening positive for posttraumatic stress disorder (PTSD) increases with severity of tinnitus [2]. We focus on tinnitus from high-frequency hearing loss that is associated with exposure to loud noise and is perpetuated by neuronal hyperactivity in the dorsal cochlear nucleus (DCN). In this study, we test the hypothesis that descending projections from the inferior colliculus (IC) cause hyperexcitability of frequencies that are no longer encoded by the bottom-up sensory signals after damage to the cochlea [3].

Methods
We implement a network model of central auditory processing that consists of 200 fusiform cells that receive tonotopic excitatory input from 200 spiral ganglion neurons (SGN) and lateral inhibitory input from 200 interneurons. We implement the descending IC projections with 200 cells that provide excitatory input to the fusiform cells. Auditory input is modeled as a depolarization of SGN neurons and hearing loss is modeled as reduced depolarization of SGN neurons at the highest frequency range. Each cell is an Izhikevich model neuron with regular spiking dynamics [4]. We characterize the dynamics of the network by applying a pure tone stimulus and simulating either normal hearing or hearing loss.
Results
Without descending IC projections, we confirm that loss of auditory nerve input at the high frequency range produces aberrant excitation at adjacent frequencies of the tonotopic map, i.e. tinnitus. With descending IC projections, we demonstrate that the signal to noise ratio increases as well as the hyperexcitability of the adjacent frequencies.
Discussion
A significant barrier to the treatment of tinnitus is the lack of knowledge on the source of the hyperexcitability; understanding the interactions between the DCN and IC in the central auditory pathway is essential to the development of physiology-based treatment to target the appropriate circuit elements. Our model shows that the descending IC projections result in hyperexcitability of high frequencies that are not encoded after hearing loss. To better understand these mechanisms, future work will involve extending the DCN model network to include narrowband and wideband inhibitors that contribute to processing pure tone, broadband, and notch noise stimuli.

Acknowledgements
This research is supported by DARPA YFA.
References
1.Annual Benefits Report 2021- Veterans Benefits Administration Reports. https://www.benefits.va.gov/REPORTS/abr/
2.Prewitt, A., Harker, G., Gilbert, T. A., et al. (2021). Mental Health Symptoms Among Veteran VA Users by Tinnitus Severity:A Population-based Survey. Military Medicine, 186(Suppl 1), 167–175.
3.Gerken, G. M. (1996). Central tinnitus and lateral inhibition: An auditory brainstem model. Hearing Research, 97(1), 75–83.
4.Izhikevich, E. M. (2003). Simple model of spiking neurons. IEEE Transactions on Neural Networks, 14(6), 1569–1572.

Speakers

Katherine Doxey

Sharon Crook

Timothy Balmer

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P072: Partial models for learning in nonstationary environments

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P072 Partial models for learning in nonstationary environments

Christopher R. Dunne*1and Dhruva V. Raman1

1Department of Informatics, University of Sussex, Falmer, United Kingdom

*Email: C.Dunne@sussex.ac.uk

Introduction
The computational goal of learning is often presented as optimality on a given task, subject to constraints. However, the notion of optimality makes strong assumptions: by definition, changes to the task structure will render an optimal agent suboptimal. This is problematic in ethological settings where an agent lacks the time or data to accurately model a task's latent states.
We present a novel design principle (Hetlearn) for learning in nonstationary environments, inspired by the Drosophila mushroom body. It makes specific predictions on what should be inferred from the environment, as compared to Bayesian inference. We show that Hetlearn outperforms an optimal Bayesian agent and better matches human and macaque behavioural data.
Methods
We consider the task of learning from reward prediction errors (RPEs) in which an animal updates a valence based on RPEs (Fig. 1 E). Critically, the degree to which the RPE changes the valence is modulated by a learning rate. To set an adaptive learning rate, Hetlearn employs parallel sublearners with heterogeneous fixed assumptions about the environment (varied fixed learning rates). Ensemble predictions employ a weighted vote with weights dependent on recent sublearner performance. This allows rapid adaptation to unpredictable environmental changes without explicitly estimating complex latent variables. We compare Hetlearn against an advanced Bayesian agent [1] and to behavioural data from humans and macaques [2, 3, 4].
Results
Hetlearn outcompetes a Bayesian agent [1] on reward learning in nonstationary environments (Fig. 1 A-D). It is also algorithmically simpler; it builds a partial generative model and does not track complex environmental statistics. Nonetheless, it aligns with behavioural data from humans and macaques as well as with previous models [2, 4] (Fig. 1 F-G). This is notable because qualitatively different models (Bayes optimal vs suboptimal) previously provided the best respective fit to these two datasets [2, 3]. As such, Hetlearn offers a unified learning principle for seemingly disparate strategies. Finally, Hetlearn is robust to model misspecification; its parameters can vary by an order of magnitude without performance decline.
Discussion
Hetlearn outcompetes [1] in part because it exploits a bottleneck in the learning process. An optimal learner needs to infer multiple quantities that impact a single bounded parameter: the learning rate. Conversely, Hetlearn tracks the recent performance of parallel learners with heterogeneous learning rates. In effect, it trades optimal performance in a stationary environment for generalisability across environments. This results in superior performance in unpredictably changing environments or those with limited time or data, which are the precise conditions in which animals outperform artificial neural networks. Crucially, Hetlearn generates new, testable predictions on what should be inferred from the environment in these regimes.

Figure 1. (A) Environments with varying statistics. (B, C) Learning rate tracking by Hetlearn and Bayesian agent [1]. (D) Hetlearn has lower mean squared error (MSE) across environments. (E) Reward prediction error (RPE) task. Bayesian agent explicitly tracks complex latent states (volatility and stochasticity) that Hetlearn tracks only implicitly. (F, G) Hetlearn matches human [2] and macaque [3, 4] data.
Acknowledgements
This research was supported by the Leverhulme Doctoral Scholarship programme be.AI – Biomimetic Embodied Artificial Intelligence at the University of Sussex.
References
[1]https://doi.org/10.1038/s41467-021-26731-9
[2]https://doi.org/10.1038/nn1954
[3]https://doi.org/10.1038/nn.3918
[4]https://doi.org/10.1016/j.neuron.2017.03.044

Speakers

Christopher Dunne

Dhruva Raman

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P073: Simulations using realistic 3D reconstructions of astrocyte endfeet reveal how cell shape alters diffusion in Alzheimer’s disease

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P073 Simulations using realistic 3D reconstructions of astrocyte endfeet reveal how cell shape alters diffusion in Alzheimer’s disease

Florian Dupeuble*1, Chris Salmon2,5, Hugues Berry1, Keith Murai2, Kaleem Siddiqi3,4,Alexandra L Schober2, J. Benjamin Kacerovsky2,Tabish A. Syed2,5,Rachel Fagen2, Tatiana Tibuleac2Amy Zhou2,Audrey Denizot1

1AIStroSight, INRIA, Université Claude Bernard Lyon 1, Villeurbanne, France
2Research Institute of the McGill University Health Centre, McGill University, Montréal, Canada
3School of Computer Science, McGill University, Montréal, Canada
4MILA - Québec AI Institute, Montreal, Canada
5Centre for Intelligent Machines, School of Computer Science, McGill University, Montreal, Canada

*Email: florian.dupeuble@inria.fr
Introduction

Astrocytes are glial cells involved in numerous brain functions, such as blood flow regulation, toxic waste clearance, or nutrient uptake [1]. They display specialized protrusions, called endfeet, that cover the majority of blood vessels and are suspected to mediate neurovascular coupling.
In Alzheimer’s Disease (AD), astrocytes undergo morphological changes [2]. However, whether endfoot morphology is altered and the functional implications of such ultrastructural changes remain poorly understood to date.

Methods
To study the impact of endfoot shape on astrocyte function, we developed a model of diffusion within high-resolution 3D reconstructions of astrocyte endfeet from WT and AD mice, derived from electron microscopy. 3D manifold tetrahedral endfoot meshes were obtained using Blender and TetWild software. Simulations of calcium diffusion were performed using FEniCS, a finite element methods Python library.
Results
We observe strong differences between the diffusional properties of AD and WT endfeet. While WT endfeet rapidly display a homogeneous calcium concentration, calcium in AD endfeet appears highly compartmentalized. Simulations accounting for the complex ER morphology suggest that they contribute to increased calcium concentration heterogeneity in endfeet, in particular in AD.
Discussion
Our preliminary results suggest that the morphological changes undergone by endfeet in AD impact local diffusion, leading to calcium compartmentalization, which could strongly affect local calcium signaling. Future work will be critical to decipher the functional link between endfoot shape, local calcium signaling, and the neurovascular uncoupling observed in AD [3]. This work provides new insights into the basic mechanisms governing endfoot dysfunction in AD.

Acknowledgements

References
1.https://doi.org/10.1146/annurev-neuro-091922-031205
2.https://doi.org/10.1016/j.coph.2015.09.011
3.https://doi.org/10.1093/brain/awac174

annurev-neuro-091922-031205

Speakers

Florian Dupeuble

Christopher Salmon

Hugues Berry

Keith Murai

Kaleem Siddiqi

Audrey Denizot

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P074: Hybridizing Machine Reinforcement Learning with Neuromimetic Navigation Systems

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P074 Hybridizing Machine Reinforcement Learning with Neuromimetic Navigation Systems

3Christopher Earl,4Moshe Tannenbaum,1Haroon Anwar,4Hananel Hazan,1,2Samuel Neymotin
1Center for Biomedical Imaging and Neuromodulation, Nathan Kline Institute for Psychiatric Research, Orangeburg, NY, USA.
2Department of Psychiatry, NYU School of Medicine, New York, NY, USA.
3Department of Computer Science, University of Massachusetts, Amherst, MA, USA.4Allen Discovery Center, Tufts University, Boston, MA, USA.
Introduction

Animal brains are capable of remembering explored locations and complex pathways to efficiently reach goals. Many neuroscience studies have investigated the cellular and circuit basis of spatial representations in the Hippocampal (HPC) and Entorhinal Cortex (EC); however, mechanisms that enable this complex navigation remain a mystery. In computer sciences, Q-Learning (QL) is a Reinforcement Learning (RL) algorithm that facilitates associations between contexts, actions, and long-term consequences. In this study, we develop a bio-inspired neuronal network model which integrates cell types from the mammalian EC and HPC hybridized with QL to simulate how the brain could learn to navigate new environments.
Methods
We used the BindsNET platform [1] to model Grid Cells (GC), Place Cells (PC), and motor control cells to drive agent actions. Our model is a Spiking Neuronal Network (SNN) with leaky integrate-and-fire (LIF) neurons, and organized to mimic GCs and PCs found in the EC and HPC (Fig 1). Reward-Modulated Spike Time Dependent Plasticity (RM-STDP) [2,3] applied to synapses activating motor control cells facilitates learning. The RM-STDP mechanism receives rewards from a Q-Table, helping the agent associate actions with long-term consequences. The agent is tasked with navigating a maze and learning a path to a goal (Fig 2). Feedback is given only at the goal, requiring the agent to associate actions with long-term outcomes to solve the maze.
Results
Trained models successfully and consistently navigated randomly generated mazes. GC populations encoded distinct physical locations into unique neural encodings, enabling the agent to distinguish between them. This lets the agent remember previously visited areas and associate them with actions. Combined with QL, long-term consequences of actions could also be retained, allowing the model to learn long paths to the goal with sparse environmental feedback.

Certain cells in the reservoir population fired only when the agent was in a specific location of the maze, suggesting these cells naturally developed PC-like characteristics. When GC’s were re-oriented in a new maze, the PC’s would remap, similar to behavior observed in biology [4].
Discussion
We designed an SNN model that mimics the mammalian brain’s spatial representation system, and integrated it with QL to solve a maze task. Our model forms the basis of a functional navigation system by effectively associating actions with long-term consequences. While the QL component is not biologically-plausible, we believe higher order brain areas could provide similar computational capabilities. In future work, we aim to implement QL as a SNN. Results also suggest an explanation for the emergence of PC in the HPC due to upstream GC activity in the EC. Moreover, GC spatial representations are likely generalizable outside of a maze. Future research could utilize our model’s GC-PC architecture to navigate more complex environments.

Figure 1. Fig 1: Diagram of bio-inspired SNN, and its relationship to QL. Bio-inspired SNN generates an action, feedback from the environment is fed into a datastructure called a ‘Q-Table’, and updates from this table modulate RM-STDP synapses in the SNN. Fig 2: Example 5x5 maze environment. Green dot represents start, blue dot the goal, yellow dot the agent position, and red dots the optimal path.
Acknowledgements
Research supported by ARL Cooperative Agreement W911NF-22-2-0139 and ARL/ORAU Fellowships
References
[1]BINDSNet: A machine learning-oriented spiking neural networks library in Python.Front Neuroinform2018 12(2018):89

[2] Training spiking neuronal networks to perform motor control using reinforcement and evolutionary learning.Front Comput Neurosci2022 16:1017284

[3] Training a spiking neuronal network model of visual-motor cortex to play a virtual racket-ball game using reinforcement learning.PLoS One2022 17(5):e0265808
[4]Remapping revisited: how the hippocampus represents different spaces.Nat Rev Neurosci202425(6):428-448

Speakers

Haroon Anwar

Christopher Earl

Hananel Hazan

Samuel A. Neymotin

Moshe Tannenbaum

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P075: Sequential dynamical invariants in winnnerless competition neural networks

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P075 Sequential dynamical invariants in winnnerless competition neural networks

Irene Elices*1, Pablo Varona1
1 Grupo de Neurocomputación Biológica, Dept. de Ingeniería Informática, Escuela Politécnica Superior, Universidad Autónoma de Madrid, 28049, Madrid, Spain.

*Email: irene.elices@uam.es
Introduction

Generating neural sequences is fundamental for behavior and cognition, which require robustness of sequential order and flexibility in its time intervals to adapt effectively. Studying cyclic sequences provides key insights into constraints limiting flexibility and shaping sequence intervals. Previously, we identified such constraints as robust cycle-by-cycle relationships between specific time intervals, i.e., dynamical invariants, in bursting central pattern generators [1]. However, their presence in computational models remains largely unexplored. Here, we examine dynamical invariants in a winnerless competition network model that generates chaotic activity while sustaining robust sequences among active neurons.

Methods
We analyzed sequence interval relationships in a Lotka-Volterra neural network that displays chaotic heteroclinic dynamics from its asymmetric connectivity [2,3]. Variables in these generalized Lotka-Volterra differential equations represent the instantaneous spike rate of the neurons. For analysis, we selected the most active neuron as a cycle reference, detecting sequential events in other neurons using activation thresholds. Cycle-by-cycle intervals were defined as the time intervals between activation and subsequent deactivation events, including those between distinct neurons. Analysis included variability measures, correlation analysis, and PCA to uncover robust relationships between interval timings.

Results
Despite the chaotic dynamics, which can be related to exploration tasks in motor and cognitive activity [2-4], we observed robust dynamical invariants between specific time intervals that added to the activation phase locks in active neurons to provide coordination between cells. The dynamical invariants represent constraints to the variability present in the chaotic activity and can underlie an emergent control mechanism. This is the first time that sequential dynamical invariants are reported in heteroclinic dynamics.

Discussion
The presence of dynamical invariants remains largely unexplored in computational models, with only a few studies addressing simplified circuits, such as minimal CPG circuit building blocks [5]. The main challenge in studying dynamical invariants in computational models is the lack of variability in individual model neurons and in network dynamics. However, a winnerless competition network model generates chaotic spatiotemporal activation patterns, thus overcoming the mentioned variability challenge. Our work analyzes for the first time the presence of dynamical invariants among the activation intervals. Results suggest that these robust cycle-by-cycle relationships are part of the sequence coordination mechanisms of the heteroclinic dynamics.

Acknowledgements
Work funded by PID2021-122347NB-I00, PID2024-155923NB-I00, and CPP2023-010818 (MCIN/AEI and ERDF- “A way of making Europe”).
References
[1]https://doi.org/10.1038/s41598-019-44953-2
[2]https://doi.org/10.1063/1.1498155
[3]https://doi.org/10.1103/PhysRevE.71.061909
[4]https://doi.org/10.1007/s11571-023-09987-3

[5]https://doi.org/10.1016/j.neucom.2024.127378

Speakers

Pablo Varona

Irene Elices Ocón

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P076: A NEST-based framework for the parallel simulation of networks of compartmental models with customizable subcellular dynamics

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P076 A NEST-based framework for the parallel simulation of networks of compartmental models with customizable subcellular dynamics

Leander Ewert1, Christophe Blaszyck2, Jakob Jordan5, Charl Linssen1,3, Pooja Babu1,3, Abigail Morrison1,2, Willem A.M. Wybo4

1Institute for Advanced Simulation (IAS-6) and JARA-BRAIN Institute I, Jülich Research Centre, 52425 Jülich, Germany
2Department of Computer Science 3 - Software Engineering, RWTH Aachen University, Aachen, Germany
3Simulation and Data Laboratory Neuroscience, Jülich Supercomputer Centre, Institute for Advanced Simulation, Jülich Research Centre, 52425 Jülich, Germany
4Peter Grünberg Institut (PGI-15), Jülich Research Centre, 52425 Jülich, Germany
5Department of Physiology, University of Bern, Bern, Switzerland

*Email: l.ewert@fz-juelich.de

Introduction

The brain is a massively parallel computer. In the human brain, 86 billion neurons convert synaptic inputs into action potential (AP) output. Moreover, even at the subcellular level, computations proceed in a massively parallel fashion. Approximately 7’000 synapse per neuron are supported by complex signaling networks within dendritic compartments. In itself, these signaling networks can also be understood as nanoscale computers that convert synaptic input, backpropagating APs, and local voltage and concentration signals into weight dynamics that support learning and memory. It is only natural, thus, to use the parallelization and vectorization capabilities of modern supercomputers to simulate the brain in a massively parallel fashion.
Methods
The NEural Simulation Tool (NEST) [1] is the reference with regards to the massively parallel simulation of spiking network models, as it has been optimized to efficiently communicate spikes across MPI processes [2]. Moreover, these capabilities introduce little overhead for the user, as the distribution of neurons across MPI processes is taken care of by NEST itself. However, so far NEST had limited options to simulate subcellular processes as part of the network, essentially forcing users to develop custom C++ codes. We have extended the scope of the NESTML modelling language [3] to support multi-compartment models, with dendrites featuring user-specified dynamical processes (Fig 1A-C).
Results
These user-specified dynamics are compiled into efficient NEST models through a C++ code generator, in such a way that the vectorization capabilities of modern CPUs are optimally leveraged. This allows for a deeper level of parallelization, next to the network parallelization across MPI processes, allowing individual CPUs to integrate up to eight compartments in parallel and decreasing single neuron runtimes accordingly. The compartmental engine furthermore leverages the Hines algorithm [4] to achieve stable and efficient integration of the system as a whole. Together, this results in single-neuron speedups compared to the field-standard NEURON simulator [5] of up to a factor of four to five (Fig 1D).
Discussion
Thus, we enable the simulation of large-scale networks where individual neurons have user-specified dynamical processes, representing (i) voltage-dependent ion channels, (ii) synaptic receptors that may be subject to a-priori arbitrary plasticity processes, or (iii) slow processes describing molecular signaling or ion concentration dynamics. Conducting such simulations has historically been challenging, since simulators specific to this purpose were lacking. With the present work, we facilitate the creation and efficient distributed simulation of such networks, thus supporting the investigation of the role of dendritic processes in network-level computations involving learning and memory.

Figure 1. Figure 1. (A) NESTML-defined subcellular mechanisms (left) are compiled into an efficient NEST model. User-defined dendritic layouts (middle) are then embedded in NEST network simulations (right). (B) NESTML code defining dendritic calcium dynamics induces BAC firing [6] in a two-compartment model [7] (C). (D) Speedup of NEST compared to NEURON (bottom) for two dendritic layouts (left vs right).
Acknowledgements
The authors gratefully acknowledge funding from the HelmHoltz POF IV, Program 2 Topic 3.

References
[1] 10.4249/scholarpedia.1430
[2] 10.3389/fninf.2014.00078
[3] 10.3389/fninf.2018.00050
[4] 10.1017/CBO9780511541612
[5] 10.1016/0020-7101(84)90008-4
[6] 10.1038/18686
[7] 10.48550/arXiv.2311.0607

Speakers

Leander Ewert

Christophe Blaszyc

Jakob Jordan

Charl Linssen

Pooja Babu

Abigail Morrison

Willem Wybo

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P077: An Intrinsic Dimension Estimator for Neural Manifolds

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P077 An Intrinsic Dimension Estimator for Neural Manifolds

Jacopo Fadanni*1, Rosalba Pacelli2, Alberto Zucchetta2, Pietro Rotondo3, Michele Allegra1,5

1Physics and Astronomy Department, University of Padova, Padova, Italy
2Istituto Nazionale di Fisica Nucleare, Sezione di Padova, Padova, Italy
3Department of Mathematical, Physical and Computer Sciences, University of Parma, Parma, Italy
4Padova Neuroscience Center, University of Padova, Padova, Italy

*Email:jacopo.fadanni@unipd.it

Introduction
Recent technical breakthroughs have enabled a rapid surge in the number of neurons that can be simultaneously recorded[1,2] calling for the development of robust methods to investigate neural activity at a population level.
In this context, it is becoming increasingly important to characterize the neural activity manifold, the set of configurations visited by the network within the Euclidean space defined by the instantaneous firing rates of all neurons[3]. A key parameter of the manifold geometry is its intrinsic dimension (ID), the number of coordinates needed to describe the manifold. While several studies suggested that the ID may be typically low, contrasting findings have disputed this statement, leading to a wide debate [1,2,3,4].
Methods
In this study we present a variant of the Full Correlation Integral (FCI), an ID estimator that was shown to be particularly robust under undersampling and high dimensionality, improving over the classical Correlation Dimension estimator[5].Our variant overcomes the limitation of standard FCI in the case of curvature effects doing a local estimation of the true IDas the peak in the distribution of local estimates. Crucially, local estimates are restricted to approximately flat neighborhoods, as determined by a suitable local parameter, which allows us to avoid overestimation. Our procedure yields a robust estimator for typically challenging situations encountered with neural manifolds.
Results
We proved the reliability of our metric by testing it in two significantly challenging cases. First, we used it to characterize neural manifolds of RNNs performing simple tasks[6], where strong curvature effects generally lead to overestimates. Second, we used it on a benchmark dataset including non-linearly embedded high-dimensional neural data, where all other methods yield underestimates[7]. In Figure 1 we show a comparison between our method and other available methods for the RNN and for the high-dimensional neural data. Linear methods overestimate the ID in the case of curved manifolds, while nonlinear methods underestimate the ID in the case of high-dimensional manifolds. In both situations, our method performed well.

Discussion
Proposing a robust estimator for the ID, our work adds a relevant tool in the open debate about the dimensionality of neural manifolds.
The intrinsic properties of the FCI estimator make it robust to undersampling and high dimensionality, avoiding the so-called ‘curse of dimensionality’ effects. Our local variant makes it robust also for curved manifolds where the ID and the embedding dimension strongly differ. Limitations of our method arise only in extremely non-uniformly sampled manifolds, where the conditions for the applicability of the FCI are unfulfilled[5].
Our method is an important step forward in the current research on neural manifolds, and it is thus of interest to the computational neuroscience community at large.

Figure 1. Left, an example of network activity projected onto the first 3 PCs. IDFCI = 2.1; IDPA = 7; MLE = 3.6. IDTwoNN = 7.1 Right: comparison between different ID estimators in the case of high-dimensional manifolds linearly embedded[7]. Our method performs well for all the dimensionality
Acknowledgements
This work was supported by PRIN grant 2022HSKLK9, CUP C53D23000740006, “Unveiling the role of low dimensional activity manifolds in biological and artificial neural networks”

References
● https://doi.org/10.1038/s41586-019-1346-5
● https://doi.org/10.1126/science.aav7893
● https://doi.org/10.1016/j.conb.2021.08.002

● https://doi.org/10.1038/s41593-019-0460-x
● https://doi.org/10.1038/s41598-019-53549-9
● https://doi.org/10.1038/s41593-018-0310-2
● https://doi.org/10.1371/journal.pcbi.1008591

Speakers

Jacopo Fadanni

Rosalba Pacelli

Alberto Zucchetta

Pietro Rotondo

Michele Allegra

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P078: Single neuron representational drift in CA1 can be explained by aligning stable low dimensional manifolds

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P078 Single neuron representational drift in CA1 can be explained by aligning stable low dimensional manifolds

Elena Faillace*1, Mary Ann Go1, Juan Álvaro Gallego1, Simon Schultz1

1Centre for Neurotechnology and Department of Bioengineering, Imperial College London, UK

*Email: elena.faillace20@imperial.ac.uk

Hippocampal place cells are believed to form a cognitive map that supports spatial navigation. However, their spatial tuning has been observed to ‘remap’, i.e. the representation drifts over time, even in the same environment [1]. This raises the question of how a robust and consistent experience is maintained despite continual remapping at the single cell level. Furthermore, it remains unclear whether this drift is coordinated across neurons, and how tuning curves profiles evolve. Here, we propose a population-level approach to identify a stable representation of environments and provide a framework to predict the activity of remapped tuning curves.

We performed two-photon calcium imaging to record the activity of hundreds of neurons in CA1 of head-fixed mice during a running task (Fig.a,b) [2]. Mice expressing GCaMP6s were habituated to a circular track for 7-9 days, followed by 3 days of recordings. All environments had the same circular structure but differed in the visual cues along the walls. During imaging, mice were exposed to two familiar environments, one novel environment, and one familiar environment with inverted order of the symbols on the walls. Neurons were longitudinally registered across sessions using CaImAn.

We used linear dimensionality reduction techniquesto find session-specific manifolds that spanned the coordinated activity of CA1 cells (Fig.c). Using a combination of PCA and canonical correlation analysis (CCA)[3,4], we were able to align these session-specific manifolds (Fig.d) across days, environments, and even mice, achieving robust decoding of the animal's position along the track (Fig.h). Moreover, using this aligned manifold, we could predict the remapping of single neuron tuning curves (Fig.e,f,g), even for those excluded when computing the alignment procedure.

This work supports the perspective that neural manifolds serve as a stable basis for neural encoding [3,4]. We present a framework in which representational drift, traditionally viewed as unstructured, can be interpreted as a coordinated adaptation at a population level, enabling the prediction of tuning curve profiles for ‘unseen’ neurons. Importantly, we did not need to categorise or select neurons based on their functional classes (e.g., place cells), thereby acknowledging their collective contribution to a preserved manifold space.

Figure 1. (a,b): schematic of experiment set-up and Ca+ imaging. Previously presented in [2]. (c): PCA of the average firing rates from different sessions concatenated. d. PCA space after each recording has been projected to a common PC space (alignment). (e,f): example of tuning curves pre and after alignment and their correlation and L2 norm (g). (h): Same as (d), colour coded by angular position.
Acknowledgements

References
[1]https://doi.org/10.1038/nn.3329
[2]https://doi.org/10.3389/fncel.2021.618658
[3]https://doi.org/10.1038/s41593-019-0555-4
[4]https://doi.org/10.1038/s41586-023-06714-0

Speakers

Elena Faillace

Mary Ann Go

Juan Alvaro Gallego

Simon R Schultz

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P079: Characterizing optimal communication in the human brain

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P079 Characterizing optimal communication in the human brain

Kayson Fakhar*1,2,Fatemeh Hadaeghi2, Caio Seguin3, Alessandra Griffa4, Shrey Dixit2,5, Kenza Fliou2,6, Arnaud Messé2, Gorka Zamora-López7,8, Bratislav Misic9, Claus Hilgetag2,10

1MRC Cognition and Brain Sciences Unit, University of Cambridge, Cambridge, UK.
2Institute of Computational Neuroscience, University Medical Center Eppendorf-Hamburg, Hamburg University, Hamburg Center of Neuroscience, Germany.
3Department of Psychological and Brain Sciences, Indiana University, Bloomington, USA.
4Leenaards Memory Center, Department of Clinical Neurosciences, Lausanne University Hospital and University of Lausanne, Montpaisible 16, 1011 Lausanne, Switzerland
5Department of Psychology, Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig, Germany.
6Sorbonne University, Paris, France.
7Center for Brain and Cognition, Pompeu Fabra University, Barcelona, Spain.
8Department of Information and Communication Technologies, Pompeu Fabra University, Barcelona, Spain.
9McConnell Brain Imaging Centre, Montréal Neurological Institute, McGill University, Montréal, Canada.
10Department of Health Sciences, Boston University, Boston, MA, USA.
*Email: kayson.fakhar@mrc-cbu.cam.ac.uk
Introduction

Efficient communication is shown to be a key characteristic of the organization of the human brain(Chen et al., 2013). In fact, it was found to bias the wiring economy of the brain networks in its favour by allocating expensive long-range shortcuts among its hubs(van den Heuvel et al., 2012). However, communication efficiency is often defined through specific signalling models — such as routing along shortest paths, broadcasting via parallel pathways, or diffusive random-walk dynamics — that omit important biological aspects of brain dynamics, including conductance delays, oscillations, and inhibitory interactions(Seguin et al., 2023). As a result, a more general framework is needed to characterize optimal signal transmission within a given brain network and to assess whether actual brain communication is truly efficient.

Methods
Here, we introduce a model-agnostic framework based on multi-site virtual lesions in large-scale neural mass models. Our approach uses a game-theoretical perspective: each brain region seeks to maximize its influence over others, subject to constraints from underlying network structure and local dynamics. This perspective yields a mathematically rigorous definition of optimal communication given any model of local dynamics on any arbitrary network structure. We used a linear, nonlinear, and oscillatory neural mass model and compared the resulting optimal influence patterns with those derived from abstract models of signalling, i.e., routing, navigation, broadcasting, and diffusion.

Results
Our results are as follows: First, we found that the broadcasting regime has the closest resemblance to the optimal communication patterns derived from game theory. Second, although the underlying structural connection weight reliably predicts the efficiency of communication between regions, it fails to capture the true influence of weakly connected hub regions. In other words, hubs harness their rich connectivity to broadcast their signal over multiple pathways when they lack a reliable direct connection to their targets. Further comparisons with functional connectivity (fMRI-based correlations) and cortico-cortical evoked potentials reveal two additional insights: (i) functional connectivity is a poor indicator of actual information exchange; and (ii) brain communication is likely to take place close to optimal levels.

Discussion
Altogether, this work provides a rigorous, versatile framework for characterizing optimal brain communication, identifies the most influential regions in the network, and offers further evidence supporting efficient signalling in the brain.

Acknowledgements

This work is in part funded by the German Research Foundation (DFG)-SFB 936-178316478-A1; TRR169-A2; SPP 2041/GO 2888/2-2 and the Templeton World Charity Foundation, Inc. (funder DOI 501100011730) under grant TWCF-2022-30510.
References
Chen, Y., Wang, S., Hilgetag, C. C., & Zhou, C. (2013). Trade-off between multiple constraints enables simultaneous formation of modules and hubs in neural systems.PLoS Comput. Biol.,9(3), e1002937.
Seguin, C., Sporns, O., & Zalesky, A. (2023). Brain network communication: Concepts, models and applications.Nat. Rev. Neurosci.,24(9), 557–574.
van den Heuvel, M. P., Kahn, R. S., Goñi, J., & Sporns, O. (2012). High-cost, high-capacity backbone for global brain communication.Proc. Natl. Acad. Sci. U. S. A.,109(28), 11372–11377.

Speakers

Kayson Fakhar

Claus Hilgetag

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P080: A Representation Learning approach captures clinical effects of slow subthalamic beta activity drifts

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P080 A Representation Learning approach captures clinical effects of slow subthalamic beta activity drifts

Salvatore Falciglia*1,2, Laura Caffi1,2,3,4, Claudio Baiata3,4, Chiara Palmisano3,4, Ioannis U. Isaias3,4, Alberto Mazzoni1,2

1The BioRobotics Institute, Scuola Superiore Sant’Anna, Pisa, Italy
2Department of Excellence in Robotics and AI, Scuola Superiore Sant’Anna, Pisa, Italy
3University Hospital Würzburg and Julius Maximilian University of Würzburg, Würzburg, Germany
4Parkinson Institute Milan, ASST G. Pini-CTO, Milan, Italy

*Email: salvatore.falciglia@santannapisa.it

Introduction

Deep brain stimulation (DBS) of the subthalamic nucleus (STN) is a mainstay treatment for drug-resistant Parkinson's disease (PD) [1]. Adaptive DBS (aDBS) dynamically adjusts stimulation according to the beta ([12-30] Hz) power of the STN local field potentials (LFPs) to match the patient's clinical status [2]. Today, aDBS control depends on accurate determination of pathological beta power thresholds [3]. Notably, in the days/months timescale, STN beta power shows irregular temporal drifts affecting the long-term efficacy of the aDBS treatment. Here we aim at characterizing these drifts and their clinical effects with a multimodal study, integrating neural and non-neural data streams.

Methods
We conducted home monitoring of patients with PD, focusing on periods of rest and gait activity. Multimodal data were collected, including STN LFPs from chronically implanted DBS electrodes, wearable inertial sensor recordings, and patient-reported diaries. A low-dimensional feature space was derived by integrating the acquired signals through Representation Learning techniques [4]. Leveraging LAURA, our transformer-based framework for predicting the long-term evolution of subthalamic beta power under aDBS therapy [5], we present a multimodal approach where neural data are paired with kinematic data and labelled according to the patient’s clinical status during the monitored activity.
Results
We observed that STN beta power distributions show large irregular non-linear fluctuations over several days. Consequently, patients spend a significant portion of time in suboptimal stimulation states. A fully informative description of the STN LFPs dynamics is achieved by integrating neural, kinematics, and clinical data into a low-dimensional feature-based representation. Latent patterns of STN activity correlate with clinical outcomes as well as motor and non-motor daily activities, necessitating further explainability within the same low-dimensional space. This might support clinically effective recalibration of aDBS parameters on a daily basis.
Discussion
Our study advances the understanding of slow timescales of pathological activity in PD patients implanted with DBS. We developed a comprehensive deep learning framework that integrates neural data with longitudinal clinical information, enabling a more precise characterization of patient status. This will enable personalized control strategies for stimulation parameters (Fig. 1) and enhance the clinician-in-the-loop paradigm by improving patient status assessment and automating aspects of neuromodulation to prevent suboptimal stimulations due to beta power drifts. Ultimately, this work paves the way for novel long-term neuromodulation strategies with potential applications to neurological disorders beyond PD [6].

Figure 1. Block diagram of aDBS as a closed-loop control system. The control loop operates on two separate timescales. In the short-term, the modulation changes with fluctuations in beta power (solid box). In the long-term, the parameters of the fast aDBS algorithm are updated based on the expected drifts of daily beta distributions combined with the neurologist’s clinical assessments (dashed box).
Acknowledgements
The authors declare that financial support was received for the research. The European Union - Next-Generation EU - NRRP M6C2 - Investment 2.1: projects IMAD23ALM MAD, Fit4MedRob, and BRIEF. Fondazione Pezzoli per la Malattia di Parkinson. The Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) Project-ID 424778381 - TRR 295.

References
1.https://doi.org/10.1007/s00221-020-05834-7
2.https://doi.org/10.1088/1741-2552/ac3267
3.https://doi.org/10.3390/bioengineering11100990
4.https://doi.org/10.1109/TPAMI.2013.50
5.https://doi.org/10.1101/2024.11.25.24317759
6.https://doi.org/10.3389/fnhum.2024.1320806

Speakers

Alberto Mazzoni

Salvatore Falciglia

Speakers

Alberto Mazzoni

Federico Fattorini

Mahboubeh Ahmadipour

Enrico Cataldo

Nicolo Meneghetti

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P085: Biological validation of a computational model of nitric oxide dynamics by emulating the nitric oxide diffusion experiment in the endothelium

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P085 Biological validation of a computational model of nitric oxide dynamics by emulating the nitric oxide diffusion experiment in the endothelium

Pablo Fernández-López1, Ylermi Cabrera-León1, Patricio García Báez1,2, Scott McElroy3, Salvador Dura-Bernal3andCarmen Paz Suárez-Araujo*1
1Instituto Universitario de Cibernética, Empresa y Sociedad, Universidad de Las Palmas de Gran Canaria, Parque Científico Tecnológico, Campus Universitario de Tafira, Las Palmas de Gran Canaria, 35017, Canary Islands, Spain.
2Departamento de Ingeniería Informática y de Sistemas, Universidad de La Laguna, Camino San Francisco de Paula, 19, Escuela Superior de Ingeniería y Tecnología, San Cristóbal de La Laguna, 38200, Canary Islands, Spain.
3State University of New York (SUNY) Downstate Health Sciences University, 450 Clarkson Avenue, Brooklyn, NY, USA 11203.

*Email: carmenpaz.suarez@ulpgc.es

Understanding how the brain works, how it is structured and how it computes is one of the goals of computational neuroscience. An essential step in this direction is to understand the cellular communication that enables the transition from nerve cells to cognition.
It is now accepted that the links between neurons are not only established by synaptic connection, but also by the confluence of different cellular signals that affect global brain activity, with the underlying mechanism being the diffusion of neuroactive substances into the extracellular space (ECS). One of these substances is the free radical gas nitric oxide (NO), which, in turn, determines a new type of information transmission: the volume transmission (VT). VT is a non-simple form of short- and long-distance communication that acts not only as a microenvironment to separate nerve cells, but also as an information channel [1, 2]. NO is a signaling molecule that is synthesized in a number of tissues by NO synthases and has the ability to regulate its own production. It is lipid soluble, membrane permeable and has a high diffusivity in both aqueous and lipid environments.
In the absence of definitive experimental data to understand how NO functions as a neuronal signalling molecule, we have developed a computational model of NO diffusion based on non-negative and compartmental dynamical systems and transport phenomena [3].
The proposed model has been validated in the biological environment, specifically in the endothelium. In this work, the biological validation is approached by reproducing the experiment performed by Tadeuzs et al, 1993 [4] on NO diffusion in the aorta. We implement our model with two compartments, using real measurements of NO synthesis and diffusion processes in the endothelial cell and in the smooth muscle cells of the aorta at a distance of 100 ± 2 µm between them. A fitting procedure to the observed NO dynamics was executed, and hypothesis related to the different processes in the NO dynamics were provided.
Our results provide evidence that the compartmental model of NO diffusion has allowed the design of a computational framework [5] to study and determine the dynamics of synthesis, diffusion and self-regulation of NO in the brain and in artificial environments. We have also shown that this model is very powerful because it allows to incorporate all the biological features and existing constraints in NO release and diffusion and in the environment where NO diffusion processes take place.
Finally, it has been shown that our model is an important tool for designing and interpreting biological experiments on the underlying processes of NO dynamics, NO behaviour and its impact on both brain structure and function and artificial neural systems.

Acknowledgements
This work has been funded by the Consejería de Vicepresidencia 1ª y de O. P., Inf., T. y M. del Cabildo de GC under Grant Nº “23/2021”, as well as by the ‘Marie Curie Chair’ under Grant Nº “38/2023”, and ‘Marie Curie Chair’ under Grant Nº “CGC/2024/9655”.The latter was funded by the Consejeria de Vicepresidencia 1ª y de Gobierno de O. P. e Inf., Arq. y V. del Cabildo de GC.
References
[1] https://doi.org/10.1177/107385849700300113.
[2] https://doi.org/10.1016/j.neuroscience.2004.06.077.
[3] https://doi.org/10.1007/978-3-319-26555-1_59.
[4] https://doi.org/10.1006/bbrc.1993.1914.
[5] https://doi.org/10.1063/1.1291268.

Speakers

Salvador Dura-Bernal

Pablo Fer

Ylermi Cabrera-León

Patricio García Báez

Scott McElroy

Carmen Paz Suárez-Araujo

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P086: Synergistic short-term synaptic plasticity mechanisms for working memory

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P086 Synergistic short-term synaptic plasticity mechanisms for working memory

Florian Fiebig*1, Nikolaos Chrysanthisdis1, Anders Lansner1,2, Pawel Herman1,3

1KTH Royal Institute of Technology, Dept of Computational Science and Technology, Stockholm, Sweden
2Stockholm University, Department of Mathematics, Stockholm, Sweden
3Digital Futures, KTH Royal Institute of Technology
*Email: fiebig@kth.se

Introduction

Working memory (WM) is essential for almost every cognitive task. The neural and synaptic mechanisms supporting the rapid encoding and maintenance of memories in diverse tasks are the subject of ongoing debate. The traditional view of WM as stationary persistent firing of selective neuronal populations has given room to newer ideas for mechanisms that support a more dynamic maintenance of multiple items, that may also tolerate activity disruption. Computational WM models based on different biologically plausible synaptic and neural plasticity mechanisms have been proposed but not combined systematically. Monolithic models (WM function explained by one particular mechanism) are theoretically appealing but also narrow explanations.
Methods
In this study we evaluate the interactions between three commonly used classes of plasticity: Intrinsic excitability (postsynaptic, increasing the excitability of spiking neurons), synaptic facilitation/augmentation (presynaptic, potentiating outgoing synapses of spiking neurons) and Hebbian plasticity (pre-post-synaptic, potentiating recurrent synapses driven by correlations), see Fig.1. Combinations of these mechanisms are systematically tested in a spiking neural network model on a broad suite of tasks or functional motifs deemed principally important for WM operation, such as one-shot encoding, free and cued recall, delay maintenance and updating. In our evaluation we focus on operational task performance and biological plausibility.
Results
We show that previously proposed short-term plasticity mechanisms may not necessarily be competing explanations, but instead yield interesting functional interactions on a wide set of WM tasks and enhance the biological plausibility of spiking neural network models. Our results indicate that a composite model, combining several commonly proposed plasticity mechanisms for WM function, is superior to more reductionist variants. Importantly, we attribute the observable differences to the principle nature of specific types of plasticity. For example, we find a previously undescribed synergistic function of Hebbian plasticity that supports the rapid updating of multi-item WM sets through rapidly learned inhibition.
Discussion
Our study suggests that commonly used forms of plasticity proposed for the buffering of WM information besides persistent activity are eminently compatible, and yield synergies that improve function and biological plausibility in a modular spiking neural network model. Combinations enable a more holistic model of WM responsive to broader task demands than what can be achieved with more reductionist models. Conversely, the targeted ablation of specific plasticity components reveals that different mechanisms are differentially important to specific aspects of WM function, advancing the search for more capable, robust and flexible models accounting for new experimental evidence of bursty and activity-silent multi-item maintenance.

Figure 1. Fig.1-Plasticity Combinations. The Augmentation plasticity model is implemented using the well-known Tsodyks-Makram mechanism [1]. The Bayesian Confidence Propagation Neural Network (BCPNN) learning rule implements intrinsic plasticity, as well as Hebbian plasticity [2]. These 3 components can be simulated separately or together, yielding 7 scenarios to simulate and study.
Acknowledgements
We would like to thank the Swedish Research Council (VR) grants: 2018-05360 and 2016-05871, Digital Futures and Swedish e-science Research Center (SeRC) for their support.
References
Tsodyks, M., Pawelzik, K., & Markram, H. (1998). Neural Networks with Dynamic Synapses.Neural Computation,10(4), 821–835.
Tully, P. J., Hennig, M. H., & Lansner, A. (2014). Synaptic and nonsynaptic plasticity approximating probabilistic inference.Frontiers in Synaptic Neuroscience,6, 8.

Speakers

Florian Fiebig

Nikolaos Chrysanthidis

Anders Lansner

Pawel Herman

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P087: Structured Inhibition and Excitation in HVC: A Biophysical Approach to Song Motor Sequences

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P087 Structured Inhibition and Excitation in HVC: A Biophysical Approach to Song Motor Sequences

Fatima H. Fneich*1, Joseph Bakarji2,Arij Daou1

1Biomedical Engineering Program, American University of Beirut, Lebanon
2Department of Mechanical Engineering, American University of Beirut, Lebanon

*Email: fhf07@mail.aub.edu

Introduction
Stereotyped neural sequencesoccurin thebrain [1], yet the neurophysiological mechanisms underlying their generation remain unclear. Birdsong is a prominent model to study such behavior, as juvenile songbirds learn from tutors andlaterproduce stereotyped song patterns. The premotor nucleus HVC coordinates motor and auditory activity for learned vocalizations.HVC consists of three neural populations with distinct in vitro and in vivo electrophysiologicalresponses [2,3]. Existing models explain HVC’s networkusingintrinsic circuitry, extrinsic feedback, or both. Here, we develop a physiologically realistic neural network model incorporating the three classes of HVC neurons basedonpharmacologicallyidentifiedion channels and synaptic currents.
Methods
We developed a conductance-based Hodgkin-Huxley-type model of HVC neurons and connected them via biologically realistic synaptic currents. The network was structured as a feedforward chain of microcircuits encoding sub-syllabic song segments, interacting through structured feedback inhibition[4]. Simulations were performed using MATLAB’s ode45 solver, incorporating key ionic currents, including T-type Ca²⁺, Ca²⁺-dependent K⁺, A-type K⁺, and hyperpolarization-activated inward current. Parameters were adjusted to replicate in vivo-like activity.The model reproduces sequential propagation of neural activity, highlighting intrinsic neuronal properties and synaptic interactions essential for song production.
Results
The model reproduced in vivo activity patterns of HVC neuron classes. HVCRA neuronsexhibitedsparse, time-locked bursts, each lasting ~10ms.HVCX neurons generated 1-4 bursts, typically following inhibitory rebound, while HVCINT neurons displayed tonic activity interspersed with bursts.Sequential propagation wasmaintainedthrough structured inhibition and excitation, with synapticconductancetuned to match dual intracellular recordings.The model accurately captured burst timing, spike shapes, and firing dynamicsobservedin experimental recordings, confirming its ability to simulate biologically realistic song-related neural activity.
Discussion
Our model provides a biophysically realistic representation of sequence generation in HVC, emphasizing the role of intrinsic properties and synaptic connectivity. The structured inhibition from HVCINT neurons ensured precise burst timing in HVCRA and HVCX neurons, supporting stable propagation. Key ionic currents, including T-type Ca²⁺ and A-type K⁺, regulated burst initiation and duration. These findings refine existing models by incorporating experimentallyobservedbiophysical details. This work offers new insights into the neural basis of motor sequence learning and could inform studies of other stereotyped behaviors.

Acknowledgements
This work was supported by the University Research Board (URB) and the Medical Practice Plan (MPP) grants at the American University of Beirut.

References
● https://doi.org/10.1038/nature09514
● https://doi.org/10.1038/nature00974
● https://doi.org/10.1152/jn.00162.2013
● https://doi.org/10.7554/eLife.105526.1

Speakers

Arij Daou

Fatima Fneich

Joseph Bakarji

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P088: Amyloid-Induced Network Resilience and Collapse in Alzheimer’s Disease: Insights from Computational Modeling

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P088 Amyloid-Induced Network Resilience and Collapse in Alzheimer’s Disease: Insights from Computational Modeling

Ediline L. F. Nguessap*1, Fernando Fagundes Ferreira1

1Department of Physics, University of São Paulo, Ribeirao Preto, Brazil

*Email: fonela@usp.br

Introduction

Alzheimer’s disease (AD) is characterized by progressive synaptic loss, neuronal dysfunction, and network disintegration due to amyloid-beta accumulation [1,2,3,4,5]. While experimental studies identifyamyloid-induced connectivity changes, the role of network resilience; the ability of the brain to maintain function despite synaptic loss remainspoorlyunderstood. Most computational models of AD either focus on static network properties (graph theory-basedapproaches) or single neuron dynamics[6], neglecting the interplay between progressive structural collapse and functional neuronal activity. Here, we model a small-world neuronal network and investigate its structural resilience and dynamical response to amyloid-driven synapse loss.

Methods
We construct a small-worldneuronalnetwork with synaptic weights evolving under amyloid-induced weakening. We track network resilience using key metrics: Largest Strongly Connected Component (LSCC) as a measure of global connectivity[7][8]. Global Efficiency, Clustering Coefficient, and Shortest Path Length to quantify functional resilience. To study functional neuronal activity, we simulate a network of Izhikevich neurons with synaptic coupling, observing how firing rates and synchronization evolve before, during, and after LSCC collapse. We further refine our model by removing isolated neurons and reducing background input when LSCC collapses, to ensure biological realism.
Results
Our simulations reveal a critical amyloid threshold (~75% synaptic loss) beyond which LSCC rapidly collapses, marking the transition from a functionally connected to a fragmented network. Small-world networks exhibit greater resilience than random ones, with LSCC persisting longer due to local clustering and efficient communication pathways. Global efficiency remains stable early on but drops sharply with LSCC collapse, while clustering initially increases (compensatory rewiring) before declining, indicating widespread disconnection. Neuronal firing desynchronizes post-collapse, aligning with cognitive dysfunction in AD, and removing isolated neurons accelerates activity decline, mimicking cortical atrophy.
Discussion
Our findings suggest that network topology plays a crucial role in Alzheimer’s resilience. As LSCC shrinks past a critical threshold, functional decline accelerates, aligning with AD progression. Neurons remain active but lose synchronization, suggesting that cortical regions stay active in late AD stages but fail to coordinate information transfer. Biologically inspired modifications (removing isolated neurons, reducing background input) enhance realism by preventing unrealistic activity after connectivity loss. This suggests that network vulnerability could serve as an AD biomarker. Future work should explore synaptic plasticity, tau pathology, and patient data (EEG, fMRI) for furtherimprovement.

Acknowledgements
FFF is supported by Brazilian National Council for Scientific and Technological Development (CNPq) 316664/2021-9. ELFN is supported by Brazilian Federal Agency for Support and Evaluation of Graduate Education (CAPES).
References

●
https://doi.org/10.1016/j.neurobiolaging.2005.10.017
●
https://doi.org/10.1212/WNL.00000000000004012
●
https://doi.org/10.1371/journal.pone.0196402
●
https://doi.org/10.1089/ars.2023.0010
●
https://doi.org/10.3389/fnbeh.2014.00106
●
https://doi.org/10.1097/NEN.0b013e31824f1c1a
●
https://doi.org/10.1016/j.amc.2021.126372
●
https://doi.org/10.1038/s41598-019-42977-6
●
https://doi.org/10.1038/nrn2575

Speakers

Ediline Laurence Fouelifack Nguessap

Fernando Ferreira

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P089: Parameter identifiability in model-based inference for neurodegenerative diseases: noninvasive stimulation

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P089 Parameter identifiability in model-based inference for neurodegenerative diseases: noninvasive stimulation

Jan Fousek*¹

¹ Central European Institute of Technology (CEITEC), Masaryk University, Brno, Czech Republic

*Email: jan.fousek@ceitec.muni.cz
Introduction

Tracking the trajectories of progression of patients with neurodegenerative diseases remains a challenging task. While employing connectome-based models can improve the performance of machine-learning-based classification [1], the identifiability of relevant parameters can be challenging when using only data features derived from spontaneous (resting state) data [2]. Here, in the context of Alzheimer disease (AD), we explore an alternative approach based on perturbations, namely using the response to single pulse transcranial magnetic stimulation recorded by EEG.

Methods
First, a whole-brain model using a normative human connectome was set up together with an EEG forward solution in order to replicate TMS evoked potential (TEP) [3] following precuneus stimulation [4]. Next, to define the trajectory of the AD in parameter space, we used previously established trajectory of progression of AD capturing how the evolution of the spatial profile of the proteinopathy expresses is reflected in the altered model parameters [5]. Using simulation-based inference, we then tried to recover the parameters using synthetic data simulated along the AD progression trajectory, and assessed the shrinkage of the posterior distributions, and the precision of the point estimates.
Results
The model successfully reproduced the TEP patterns found in the empirical data. Along the progression trajectory, the model parameters remained identifiable, showing significant shrinkage of the posterior distribution with respect to the prior and small distance of the mean values from the ground-truth. Additionally, while we observed some correlation between the estimated parameters (hinting to a certain degree of degeneracy), it did not impact the performance of the inference.
Discussion
Here we demonstrate that the brain response to the noninvasive stimulation is informative enough to allow effective parameter inference in connectome-based models. The workflow can be easily adapted to different data-features derived from the TEPs, as well as different stimulation targets. As a natural next step, this approach will be benchmarked and validated in empirical datasets on individual subject data.

Acknowledgements
Jan Fousek receives funding from the European Union’s Horizon Europe research and innovation programme under the Marie Sklodowska-Curie grant agreement No 101130827.
References
[1] https://doi.org/10.1002/trc2.12303
[2] https://doi.org/10.1088/2632-2153/ad6230
[3] https://doi.org/10.3389/fninf.2013.00010
[4] https://doi.org/10.1016/j.clinph.2024.09.007
[5] https://doi.org/10.1523/ENEURO.0345-23.2023

Speakers

Jan Fousek

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P090: Identifying Manifold Degeneracy and Estimating Confidence for Parameters of Compartmental Neuron Models with Hodgkin-Huxley Type Conductances

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P090 Identifying Manifold Degeneracy and Estimating Confidence for Parameters of Compartmental Neuron Models with Hodgkin-Huxley Type Conductances

Saina Namazifard1, Anwar Khaddaj2, Matthias Heinkenschloss2,Fabrizio Gabbiani*1

1Department of Neuroscience, Baylor College of Medicine, Houston, USA
2Department of Computation Applied Mathematics & Operations Research, Rice University, Houston, USA
*Email:gabbiani@bcm.edu

Introduction

Much work has been devoted to fitting the biophysical properties of neurons in compartmental models with Hodgkin-Huxley type conductances. Yet, little is known on how reliable model parameters are, and their possible degeneracy. For example, when characterizing a membrane conductance through voltage-clamp (VC) experiments, one would like to know if the data will constrain the parameters and how reliable their estimates are. Similarly, when studying the responses of a neuron with multiple conductances in current clamp (CC), how robust is the model to changes in peak conductances. Such degeneracy is linked to biological robustness [1] and is key in understanding the constraints posed by conductance distributions on dendritic computation [2].

Methods
A one-compartment model with Hodgkin-Huxley (HH) type conductances was used. We studied synthetic and experimental VC data of the H-type conductance (gH) that is widely expressed in neuronal dendrites. We also studied the original HH model in VC and CC. Finally, we considered a stomatogastric ganglion (STG) neuron model in CC. The ordinary differential equation solutions, parameters, and their sensitivities were simultaneously estimated using collocation methods and automatic differentiation. This allowed to solve the non-linear least squares (NLLS) problem associated with each model. Parameter degeneracy manifold iterative tracing was performed based on the singular value decomposition (SVD) of the NLLS residual Jacobian.
Results &Discussion
We identified parameter degeneracy using an SVD-based subset selection algorithm [3] applied to the objective function Jacobian. In the gH model in VC, the 2 least identifiable parameters were the leak conductance (gL) and gHreversal potentials, ELand EH. EHwas constrained by tail current experiments. This left a 1-dimensional (1-D) non-linear solution manifold for the remaining 7 parameters: gL, EH, and peak gHat 5 VC values. In the HH model in VC, 3 parameters were least identifiable: EK, gNaand EL. The HH model in CC exhibited approximate parameter degeneracy with a 1-D solution manifold. Similar results were obtained for the STG model. The role of ELin degeneracy was unexpected. Our results generalize to multi-compartment models.

Acknowledgements
Supported by NIH grant R01 NS130917.

References
1. Marom, S., & Marder, E. (2023). A biophysical perspective on the resilience of neuronal excitability across timescales.Nature Reviews Neuroscience,24, 640–652.https://doi.org/10.1038/s41583-023-00730-9
2. Dewell, R. B., Zhu Y., Eisenbrandt M., Morse R., & Gabbiani F. (2022). Contrast polarity-specific mapping improves efficience of neuronal computation for collision detection. Elife, 11:e79772.https://doi.org/10.7554/eLife.79772
3. Golub, G. H., & Van Loan C. F. (2013). Matrix Computations (4thed.). John Hopkins University Press.https://epubs.siam.org/doi/book/10.1137/1.9781421407944

Speakers

Fabrizio Gabbiani

Saina Namazifard

Anwar Khaddaj

Matthias Heinkenschloss

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P091: Dynamic causal modelling (DCM) for effective connectivity in MEG High-Gamma-Activity: a data-driven Markov-Chain Monte Carlo approach

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P091 Dynamic causal modelling (DCM) for effective connectivity in MEG High-Gamma-Activity: a data-driven Markov-Chain Monte Carlo approach
P. García-Rodríguez, M. Gilson, J.-D. Lemaréchal, A. Brovelli

CNRS UMR 7289 - Aix Marseille Université, Institut de Neurosciences de la Timone, Campus Santé Timone, Marseille, France

Email: pedro.garcia-rodriguez@univ-amu.fr

Introduction

Model inversion in DCM traditionally considers the application of Bayesian variational schemes,i.e.quadratic approximations in the vicinity of minima in the parameters space [1]. On the other hand, more general Markov Chain Monte Carlo methods (MCMC) opt for an intense use of random numbers to sampling posterior probability distributions. The successful application of either of them highly depend on the correct choice of prior distributions.

Methods

Here we propose an automated workflow combining MCMC with more conventional optimization Gradient-Descent (GD) techniques. Following the bi-linear model [2], a simpler DCM is considered with a matrixAforeffective connectivity and a matrixCfor sensory driving inputs.AlphaandGammafunctions for input profiles complete the modeling scenario.

Model’s parameters are estimated in three parts.Firstly, the matrixAis initialized from a Gaussian distribution with null-mean and variance given by observer-specific or group-level Granger Causality (GC) computed from the data. Next, GD algorithms implement a constraint bounded optimization to keep input parameters within plausible (positive) intervals. The adequacy of the parameters values found are further tested through a Levenberg-Marquard GD form. Finally, a MCMC Bayesian scheme incorporates the covariance of the observation noise in a Multivariate Gaussianlikelihood model. A Generative Model is so completed with parameter's prior distributions based on the GD optimizations mentioned above. Normal or Log-Normal distributions are alternatively used, the later to ensure positive values after sampling when needed.

Results

The approach is applied to High-Gamma Activity induced responses during visuomotor transformation tasks executed by 8 subjects, as reported in [3]. Methods were applied to hundred of trials for each subject, providing a handy data-driven DCM framework to evaluate the plausibility of various model configurations. Observation noise is empirically estimated from the pre-stimulus periods in original trials. Model inversion pipeline tends to support the most realistic model configuration tested, with an apparent relation between the estimated effective connectivityAandmatrixGC(Fig. 1).

Discussion
Comparison of prior and posterior distributions can help distinguish informative from non-informative
parameters.Initialization of matrixAwith structural connectivity instead was tested.

Figure 1. A DCM for high-gamma-activity (HGA). First column: brain regions and model configurations tested (top) and corresponding Granger-Causality (GC) matrix (bottom). Second column: model predictions compared to experimental HGA profiles (top) and relation between GC and estimated effective connectivty matrix A (bottom).
Acknowledgements
A.B. and P.G-R were supported by EU’s Horizon 2020 Framework Programme for Research and Innovation under the Specific Grant Agreements No. 101147319 (EBRAINS 2.0 Project).
References

[1] Zeidman P, Friston K, Parr T. (2023) A primer on Variational Laplace (VL). Neuroimage 279:120310. doi: 10.1016/j.neuroimage.2023.120310.[2] Chen CC, Kiebel SJ, Friston KJ (2008). Dynamic causal modelling of induced responses. Neuroimage 41(4):1293-1312. DOI: 10.1016/j.neuroimage.2008.03.026. PMID: 18485744.

[3] Brovelli A., Chicharro D., Badier J-M., Wang H., Jirsa V. (2015). Characterization of Cortical Networks and Corticocortical Functional Connectivity Mediating Arbitrary Visuomotor Mapping. J. Neuroscience 35(37):12643-12658. doi: 10.1523/JNEUROSCI.4892-14.2015.

Speakers

Matthieu Gilson

chair of junior professor, Aix-Marseille University

Pedro García-Rodríguez

Jean-Didier Lemarechal

Andrea Brovelli

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P092: Reversal of Wave Direction in Unidirectionally Coupled Oscillator Chains

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P092 Reversal of Wave Direction in Unidirectionally Coupled Oscillator Chains

Richard Gast*1, Guy Elisha2, Sara A. Solla3,4 , Neelesh A. Patankar5

1Department of Neuroscience, The Scripps Research Institute, San Diego, US
2Brain Mind Institute, EPFL, Lausanne, Switzerland
3Department of Neuroscience, Northwestern University, Evanston, US
4Department of Physics and Astronomy, Northwestern University, Evanston, US
5Department of Mechanical Engineering, Northwestern University, Evanston, US

*Email: rgast@scripps.edu

Introduction:Chains of coupled oscillators have been used to model animal behavior such as crawling, swimming, and peristalsis [1]. In such chains, phase lags between adjacent oscillators yield a propagating wave, which can either be anterograde (from proximal to distal) or retrograde (from distal to proximal). Switches in the direction of wave propagation have been related to increased flexibility, but also to pathology in biological systems. In Drosophila larvae, for example, switches in wave propagation are required for crawling, which has been achieved in a coupled oscillator chain model by applying an extrinsic input to distinct ends of the chain [2].

Methods:In this work, we explore a different, novel mechanism for reversing the wave propagation direction in a chain of unidirectionally coupled limit cycle oscillators. Instead of requiring tuned coupling or precisely timed local inputs, changes in the global extrinsic drive to the chain of oscillators suffices to control the direction of wave propagation. To this end, we consider a chain of unidirectionally coupled Wilson-Cowan (WC) oscillators [3]. The system is driven bySEandSI, which are extrinsic inputs globally applied to all excitatory and inhibitory populations in the chain, respectively.

Results:Combining numerical simulations and bifurcation analysis, we show that waves can propagate in anterograde or retrograde directions in the unidirectional chain of WC oscillators, despite uniform coupling and extrinsic input strengths across the chain [4]. We find that the direction of propagation is controlled by a disparity between the intrinsic frequency of the proximal oscillator and that of the more distal oscillators in the chain (see figures in [4]). The transition between these two behaviors finds explanation in the proximity of the chain's operational regime to a homoclinic bifurcation point, where small changes in the input translate to strong shifts in the oscillation period.

Discussion:Lastly, we discuss wave propagation in the context of phase oscillator networks. We describe a direct relationship between the intrinsic frequency differences between the proximal and distal chain elements, and the phase shift parameter of a phase coupling function [4]. This way, we analytically extend our numerical results to a more general phase oscillator model. Our work emphasizes the functional role that the existence of a homoclinic bifurcation plays for activity propagation in neural systems. The ability of this mechanism to operate on time scales as fast as the neural activity itself suggests that it could dynamically emerge in a variety of biological systems.

Acknowledgements
This work was funded by the by the National Institutes of Health (NIDDK Grant
No. DK079902 and No. DK117824), and National Science Foundation (OAC Grant No. 1931372).
References
[1] Kopell, N., & Ermentrout, G. B. (2003).The Handbook of Brain Theory and Neural Networks.

[2] Gjorgjieva, J., Berni, J., Evers, J. F., & Eglen, S. J. (2013).Frontiers in Computational Neuroscience, 7, 24.

[3] Wilson, H. R., & Cowan, J. D. (1972).Biophysical journal, 12(1), 1-24.
[4] Elisha, G., Gast, R., Halder, S., Solla, S. A., Kahrilas, P. J., Pandolfino, J. E., & Patankar, N. A. (2025).Physical Review Letters, 134(5), 058401.

Speakers

Richard Gast

Camille Godin

Jean-Philippe Thivierge

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P098: DelGrad: Exact event-based gradients in spiking networks for training delays and weights

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P098 DelGrad: Exact event-based gradients in spiking networks for training delays and weights

Julian Göltz*+1,2, Jimmy Weber*3, Laura Kriener*3,2,
Sebastian Billaudelle3,1, Peter Lake1, Johannes Schemmel1,
Melika Payvand$3, Mihai A. Petrovici$2

1Kirchhoff-Institute for Physics, Heidelberg University, Heidelberg, Germany
2Department of Physiology, University of Bern, Bern, Switzerland
3Institute of Neuroinformatics, University of Zurich and ETH Zurich, Zurich, Switzerland

*Shared first authorship$shared senior authorship
+Email: julian.goeltz@kip.uni-heidelberg.de
Introduction
Spiking neural networks (SNNs) inherently rely on the timing of signals for representing and processing information. Incorporating trainable transmission delays, alongside synaptic weights, is crucial for shaping these temporal dynamics. While recent methods have shown the benefits of training delays and weights in terms of accuracy and memory efficiency, they rely on discrete time, approximate gradients, and full access to internal variables like membrane potentials [1]. This limits their precision, efficiency, and suitability for neuromorphic hardware due to increased memory requirements and I/O bandwidth demands.

Methods
To alleviate these issues, building on prior work on exact gradients in SNNs [2] we propose an analytical approach for calculating exact gradients of the loss with respect to both synaptic weights and delays in an event-based fashion. The inclusion of delays emerges naturally within our proposed formalism, enriching the model’s parameter search space with a temporal dimension (Fig. 1a). Our algorithm is purely based on the timing of individual spikes and does not require access to other variables such as membrane potentials.

Results
We investigate the impact of delays on accuracy and parameter efficiency both in ideal and hardware-aware simulations on the Yin-Yang classification task [3]. Furthermore, while previous work on learnable delays in SNNs has been mostly confined to software simulations, we demonstrate the functionality and benefits of our approach on the BrainScaleS-2 neuromorphic platform [4, Fig. 1b], successfully training on-chip-delays, and showing a good correspondence to our hardware-aware simulations (Fig. 1c,d).

Discussion
DelGrad presents an event-based framework for gradient-based co-training of weight and delay parameters, without any approximations. For the first time, we experimentally demonstrate the memory efficiency and accuracy benefits of adding delays to SNNs on noisy mixed-signal hardware. Additionally, these experiments also reveal the potential of delays for stabilizing networks against noise. DelGrad
opens a new way for training SNNs with delays on neuromorphic hardware, which results in fewer required parameters, higher accuracy and ease of hardware training.

Figure 1. a Information flow in an SNN, effect of weights w and delays d on the membrane potential of a neuron, and raster plot of the activity. b Photo of the neuromorphic chip BrainScaleS-2. c Comparison of networks without (blue) and with (orange) delays, showing the benefit of delays. d Our hardware-aware simulation can be used effectively as a proxy for hardware emulation, and confirms these benefits.
AcknowledgementsThis work was funded by the Manfred Stärk Foundation, the EC Horizon 2020 Framework Programme under grant agreement 945539 (HBP) and Horizon Europe grant agreement 101147319 (EBRAINS 2.0), the DFG under Germany’s Excellence Strategy EXC 2181/1-390900948 (STRUCTURES Excellence Cluster), SNSF Starting Grant Project UNITE (TMSGI2-211461), and the VolkswagenStiftung under grant number 9C840.
References
[1] I. Hammouamri, et al. doi: 10.48550/arXiv.2306.17670.
[2] J. Göltz, et al. doi: 10.1038/s42256-021-00388-x.
[3] L. Kriener, et al. doi: 10.1145/3517343.3517380.
[4] C. Pehle, et al. doi: 10.3389/fnins.2022.795876.

Speakers

Mihai A. Petrovici

Julian Goeltz

Jimmy Weber

Laura Kriener

Sebastian Billaudelle

Peter Lake

Johannes Schemmel

Melika Payvand

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P099: Critical Slowing Down and the Hierarchy of Neural Timescales: A Unified Framework

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P099 Critical Slowing Down and the Hierarchy of Neural Timescales: A Unified Framework

Leonardo L. Gollo*1,2

1Brain Networks and Modelling Laboratory and The Turner Institute for Brain and Mental Health, Monash University, Melbourne, Australia.
2Instituto de Física Interdisciplinar y Sistemas Complejos, IFISC (UIB-CSIC), Palma de Mallorca, Spain

*Email: leonardo@ifisc.uib-csic.es

IntroductionResearch on brain criticality often focuses on identifying phase transitions, typically assuming that brain dynamics can be described by a single control parameter [1]. However, this approach overlooks the inherent heterogeneity of neural systems. At the neuronal level, diversity in excitability gives rise to multiple percolations and transitions, leading to complex dynamical behaviors [2]. At the macroscopic level, this heterogeneity enables the brain to operate across a broad hierarchy of timescales [3], ranging from rapid neural responses to external stimuli to slower cognitive processes [4,5]. A critical open question is how the framework of brain criticality, which emphasizes phase transitions, can be reconciled with the observed hierarchy of neural timescales.
MethodsWe employed a theoretical framework integrating nonlinear dynamics and criticality theory to analyze the relationship between hierarchical timescales and proximity to criticality. Specifically, we examined the role of critical slowing down, a phenomenon in which systems near a phase transition exhibit prolonged recovery times following perturbations. Using existing empirical findings on functional brain hierarchy and criticality [6,7,8], we evaluated how regions with slower timescales align with the principles of critical slowing down. Additionally, we explored how this framework supports a balance between sensitivity and stability in neural information processing [9].
ResultsOur analysis indicates that brain regions are not uniformly critical but instead positioned at varying distances from criticality. Regions with slower timescales tend to be situated closer to the critical point due to critical slowing down, while regions with faster dynamics operate in subcritical regimes. This spatiotemporal organization supports a structured coexistence of critical and subcritical dynamics, which enhances both sensitivity to external stimuli and reliable internal processing. Furthermore, this framework naturally gives rise to a hierarchy of timescales, and the coexistence of critical and subcritical dynamics enables a balance between flexibility and robustness, allowing neural systems to dynamically regulate information flow and cognitive processes [9].
Discussion
By integrating brain criticality and hierarchical timescales, our findings offer a novel perspective on neural dynamics. Instead of a uniform critical state, we propose that brain regions exist on a criticality continuum, shaped by their functional roles and temporal properties. This unified framework provides a nonlinear dynamics explanation for the brain’s timescale-based hierarchy, shedding light on its neurophysiological mechanisms. By bridging criticality and hierarchical organization, this work advances our understanding of the fundamental principles governing brain dynamics, offering a foundation for future investigations into neural computation and cognition.

Acknowledgements
We thank our colleagues and collaborators for their insightful discussions and feedback, which have enriched the development of this work. This work was supported by the Australian Research Council (ARC) Future Fellowship (FT200100942), the Ramón y Cajal Fellowship (RYC2022-035106-I), and the María de Maeztu Program for units of Excellence in R&D, grant CEX2021-001164-M/10.13039/501100011033.
References1.https://doi.org/10.1016/j.pneurobio.2017.07.002
2.https://doi.org/10.7717/peerj.1912
3.https://doi.org/10.1371/journal.pcbi.1000209
4.https://doi.org/10.1098/rstb.2014.0165
5.https://doi.org/10.1523/JNEUROSCI.1699-24.2024
6.https://doi.org/10.1073/pnas.2208998120
7.https://doi.org/10.1371/journal.pcbi.1010919
8.https://doi.org/10.1103/PhysRevX.14.031021
9.https://doi.org/10.1098/rsif.2017.0207

Speakers

Leonardo Gollo

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P100: Dynamic Range Revisited: Novel Methods for Accurate Characterization of Complex Response Functions

P103: Developmental Changes in Circuit Dynamics during Hippocampal Memory Recall

P106: Computational and Experimental Insights into Hippocampal Slice Spiking under Extracellular Stimulation

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P106 Computational and Experimental Insights into Hippocampal Slice Spiking under Extracellular Stimulation

Sarah Hamdi Cherif*1,2,3, Mathilde Wullen4, Steven Le Cam2, Valentine Bouet4, Jean-Marie Billard4, Jérémie Gaidamour3, Laure Buhry1 , Radu Ranta2

1Université de Lorraine, CNRS, LORIA, France
2Université de Lorraine, CNRS, CRAN, France
3Université de Lorraine, CNRS, IECL, France
4 Normandie Univ, UNICAEN, CHU Caen, INSERM, CYCERON, COMETE, France

*Email: sarah.hamdi-cherif@loria.fr

Introduction

Synaptic plasticity and neuronal excitability in the hippocampus (HC) are altered in schizophrenia [1]. Multi Electrode Array (MEA) recordings following a long-term potentiation (LTP) protocol revealed local field potential (LFP) variations along physiological pathways and high-frequency (HF) activity near the stimulation site [2]. To understand the effect of extracellular stimulation (ES) and explore its relationship with synaptic activity and spike generation, we combined electrophysiological recordings and computational modelling. We applied ES to hippocampal slices around Schaeffer’s collaterals while recording signals near CA3 pyramidal cell bodies, and we developed a computational model to aid interpretation.
Methods
Experiments: in-depth glass microelectrode recordings in CA3 of healthy HC slices. 40 pulses (0.4 ms duration, 7 s apart) at 0.2, 0.4 mA. Signals were processed and filtered above 300 Hz to isolate spiking activity.
Simulations: one multi-compartment CA3 pyramidal neuron model [3], with ES modelled using LFPy as a dipole [4], orthogonal to Schaffer collaterals. Background noise below spiking level was added to reflect environmental HC conditions. We varied the position of the stimulation along the axon and at different distances to explore spike variability. Synaptic inputs included excitatory (dendritic) and inhibitory (somatic) drive [5], generated by a variable-rate Poisson process, simulating activation of cells located closer to the ES.
Results
In the experimental data, each ES pulse triggered a single spike, issued from the same cell according to tentative spike sorting [6]. Lower pulse intensity led to variable latencies. As intensity increased, spike timing arose earlier and became more synchronized(Fig.1.a).

According to our simulations (Fig.1.b), a cell, activated directly by ES, showed spike latencies of 0.25–4 ms that were used to parametrize the Poisson process. The target cell, excited both by ES and synaptic inputs, exhibited later and more dispersed latencies, of 3-8 ms, closer to experimental data, suggesting they capture further-layer activity. Higher intensity had the same effects as experimental data (Fig.1.c).
Discussion
Our findings suggest that the HF activity observed in the MEA recordings results from spiking activity propagating antidromically within CA3, activating recurrent excitatory networks. We plan more recordings to confirm our findings and use the model to include the reproduction of LFP dynamics, focusing on synaptic activity. Also, to work on simulating populations while reducing the cell-models to point-neurons and implementing a network parametrized using the observed spike latencies to approximate its dynamics. Ultimately, we aim to develop a comprehensive computational model of HC electrical activity and synaptic plasticity, in healthy and schizophrenia mouse models [7], to better understand the mechanisms involved.

Figure 1. Figure 1 - (a) Experimental trials at 200µA (left) and 400µA (right), (b) Simulated membrane potentials at 100µA (left) and 250µA (right). Both trials and simulations last 50 ms, with stimulation starting at 5 ms. (c) Boxplots comparing spike latency variability across low/high intensities in both experiment (left) and simulation (right).
Acknowledgements
//
References
[1]https://doi.org/10.3390/ijms22052644
[2]https://doi.org/10.12751/nncn.bc2024.244
[3]https://doi.org/10.1523/JNEUROSCI.1889-24.2025
[4]https://doi.org/10.1007/978-1-61779-170-3_8
[5]https://doi.org/0.3389/fncel.2013.00262.
[6]https://doi.org/10.1088/1741-2552/acc210
[7]https://doi.org/10.1016/j.schres.2020.11.043

Speakers

Sarah Hamdi Cherif

Laure Buhry

Mathilde Wullen

Steven Le Cam

Valentine Bouet

Jean-Marie Billard

Jérémie Gaidamour

Radu Ranta

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P107: Towards A New Metric for Understanding Neural Population Encoding in Early Vision

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P107 Towards A New Metric for Understanding Neural Population Encoding in Early Vision

Silviya Hasana*1, Simo Vanni1

1Department of Physiology, Medicum, University of Helsinki, Helsinki, Finland

*Email: silviya.hasana@helsinki.fi

Introduction
Representational fidelityofvisioncan be evaluatedusing adecoding approach[1-3],howeverthe methodisinexplicableand difficult to quantify.This study aims to develop a quantitative metric for vision models based on neural population spike activity.By analyzing the relationship between stimulus features, spike timing, and receptive field locations, we investigate how population spike data encode information.We apply both deterministic and probabilistic approaches to evaluate neural populationsencoding,andsubsequentlyquantitativedecodingcapacityonspatialstimulus information.Aquantitativeperformancemetricis fundamental for advancing functional computational vision models, such as the SMART system[4].

Methods
We useda macaque retina model with simulatedON and OFFparasolunitsin a 2D retinal patch.Stimuli were varied for spatial parameters, such asspatial frequency and orientation.First, we summed the spikes generated within 500msof grating stimulus onset for each unit separately.Then, wecalculated the difference between ON- and OFF-unit spike counts and normalized the responses.Toevaluatewhether activation patterns aligned with thetruestimulus, we binned the responses and applied a deterministic Gabor filteratdifferent orientations.Subsequently, we plan to evaluatemodel performanceusing a Bayesian Ideal Observer, which models prior, likelihood, and posterior as a tuning curve foroptimalstimulus decoding.

Results
Our findings showed the presence of orientation-specific patterns in neural population activity, both in the deterministic and probabilistic approaches. Based on preliminary data analysis and processing in our deterministic approach, we expected a strong match between Gabor kernel prediction and true orientations for parasol ON and OFF cells. Our experimentstested100 sweeps, obtained 100% accuracy for oblique orientation prediction with mean average errorof2.97degrees. The high accuracy of the deterministic approach confirms that simple feature-based encoding mechanisms, such as Gaborfiltermatching, align well with neural responses in the parasol ON and OFF cells.

Discussion
As expected, the modeled retinal ganglion cell population encodes orientation in a structured manner that can be decoded based on receptive field positions in the visual field.Moving forward, we will explorea probabilistic approach by applying curve tuning through a Bayesian Ideal Observer to assessthe reliability of neuron population spike activityencodes stimulus orientation, spatial frequency, and motion direction.The probabilistic approach will incorporate prior and likelihood to reconstruct stimulus orientation. The resultswillassesshowboth deterministic andprobabilisticapproaches complement, andcontributetoneuraldecoding, providing a quantitative metric for evaluating functional vision models.

AcknowledgementsThis work has been supported by Academy of Finland grant No: 361816
References
[1]https://doi.org/10.1371/journal.pcbi.1006897
[2]https://doi.org/10.1038/s41583-021-00502-3
[3]https://doi.org/10.1038/nrn2578
[4]https://doi.org/10.1016/j.brainres.2008.04.024

Speakers

Silviya Hasana

Simo Vanni

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P109: From point neurons to biophysically detailed networks: A data-driven framework for multi-scale modeling of brain circuits

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P109 From point neurons to biophysically detailed networks: A data-driven framework for multi-scale modeling of brain circuits

Beatriz Herrera*1, Xiao-Ping Liu2, Shinya Ito1, Darrell Haufler1Kael Dai1, Brian Kalmbach2, Anton Arkhipov1
1Allen Institute, Seattle WA, 98109, USA
2Allen Institute for Brain Science, Seattle WA, 98109, USA

*Email: beatriz.herrera@alleninstitute.org
Introduction
Patch-seq technique links transcriptomics, neuron morphology and electrophysiology in individual neurons. Recent work at the Allen Institute resulted in a comprehensive mouse and human neuron database with Patch-seq data for diverse cell types. In this study, we carry out a large-scale optimization of generalized-leaky integrate-and-fire (GLIF) models on these Patch-seq data for thousands of cells. Furthermore, anticipating applications of models of diverse cell types for network simulations at multiple levels of resolution, we propose a strategy to convert point-neuron network models into detailed biophysical models.Methods
GLIF modelswere obtained from the Patch-seq electrophysiology recordings, and we quantified the optimization performance for different types of current injection stimulus, as well as comparing with an earlier approach[1].
GLIF biophysical conversion strategyinvolved (1) mapping each GLIF neuron to a biophysical neuron model; (2) replacing GLIF network parameters with corresponding biophysical parameters[2]; (3) estimating conversion factors to translate current-based synaptic weights to conductance-based for each source-target pair; (4) constructing the biophysical network model using the scaling factors from (3).Results
We find that optimizing GLIF models using long-square step current stimuli generalizes better to noise stimuli (than vice versa). With this approach, we obtained GLIF models for a total of 6,460 cells from diverse types of both mouse and human glutamatergic and GABAergic interneurons[3–6].
We tested our GLIF-to-biophysical network conversion on our V1 point-neuron model[2]. We simulated responses to pre-synaptic populations and calculated synaptic weight factors for matching GLIF firing rates. We built the V1 biophysical model, fine-tuning weights to align with recordings and validating againstin vivoNeuropixels data.Discussion
Our work establishes the foundation for more comprehensive simulations of brain networks. We shed light on the relationships between genes and morpho-electrophysiological features by developing models for various cell types with available transcriptomic data from Patch-Seq experiments. Furthermore, our method for transforming point-neuron network models into detailed biophysical models will aid in developing and optimizing such complex models, as point-neuron networks are less computationally intensive and simpler to optimize for reproducing experimental data.

Acknowledgements
We thank the founder of the Allen Institute, Paul G. Allen, for his vision, encouragement, and support. This work was supported by the National Institutes of Health (NIH) under the following award nos.: NIBIB R01EB029813, NINDS R01NS122742 and U24NS124001, and NIMH U01MH130907. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.
References
1. Teeter C, Iyer R, Menon V, et al. Nature Communications 2018; 9:
2. Billeh YN, Cai B, Gratiy SL, et al. Neuron 2020; 106:388-403.e18
3. Berg J, Sorensen SA, Ting JT, et al. Nature 2021; 598:151–158
4. Gouwens NW, Sorensen SA, Baftizadeh F, et al. Cell 2020; 183:935-953.e19
5. Chartrand T, Dalley R, Close J, et al. Science 2023; 382:eadf0805
6. Lee BR, Dalley R, Miller JA, et al. Science 2023; 382:eadf6484

Speakers

Beatriz Herrera

Anton Arkhipov

Xiao-Ping Liu

Shinya Ito

Darrell Haufler

Kael Dai

Brian Kalmbach

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P110: What is membrane electrical excitability?

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P110 What is membrane electrical excitability?

Lionardo Truqui2, Hector Chaparro-Reza1, Vania Victoria Villegas1, Marco Arieli Herrera-Valdez1*

1Dinamics, Biophysics, and Systems Physiology Laboratory, Universidad Nacional Autónoma de México.
2Graduate Program in Mathematics, Universidad Nacional Autónoma de México

*Email: marcoh@ciencias.unam.mx

Neuronal excitability is a phenomenon that is understood in different ways. A neuron may be regarded as more excitable than another if it responds to a stimulus with more action potentials within a fixed period of time. Another way to think about how excitable a neuron is could be to consider the delay with which it starts to respond to a given stimulus. We use the simplest, 2-dimensional, biophysical model of neuronal membrane potential based on two transmembrane currents carried by sodium and potassium ions similar to the Morris-Lecar model[4], but without leak current [1], to study conditions that should be satisfied by an excitable system, and provide a formal definition of electrical excitability. The model consist only on two currents, a Na and a K current, as small currents are not necessary to generate action potentials [3]. We first establish the notion that a model based on autonomous evolution rules is associated with a family of dynamical systems. For instance, if the parameter representing the input current in the equation for the membrane potential is varied todescribe experimental data in current-clamp experiments, the family is defined at least by the input current, and its members can be associated to different sets of trajectories in phase space. We then proceed to analyse the properties of single dynamical systems by examination of their underlying vector fields. In a similar way as originally proposed by Fitz-Hugh [2], we define a region from which all trajectories are action potentials, and call it the Excitability Region. We also propose a measure to quantify the extent to which a single dynamical system is excitable, and then proceed to compare different degrees of excitability. Since the membrane potential of a neuron is represented by a family of dynamical systems, we then examine which of those systems are excitable under the above definition, and assess which ones are more excitable, as a function of the input current. While doing so, we explore the bifurcation structure of the model taking the input current as the bifurcation parameter, and characterize the changes in excitability induced by varying the sizes of the population of ion channels. Having done so, we define neuronal excitability by extending our definition for a single dynamical system to the whole family in the model. We discuss how our measure of excitability behaves around attractor nodes and attractor foci, and also use our definitions to describe the I-F relations of types I, II, and III, that have been used previously to characterize excitability.

Acknowledgements
Universidad Nacional Autónoma de México
References
[1] Av-Ron, E., Parnas, H., and Segel, L. A. (1991). A minimal biophysical model for an excitable and oscillatory neuron. Biological Cybernetics, 65(6):487–500.
[2] FitzHugh, R. (1961). Impulses and physiological states in theoretical models of nerve membrane. Biophysical journal, 1(6):445–466.
[3] Herrera-Valdez, M. A. and Lega, J. (2011). Reduced models for the pacemaker dynamics of cardiac cells. Journal of Theoretical Biology, 270(1):164–176.
[4] Morris, C. and Lecar, H. (1981). Voltage oscillations in the barnacle giant muscle fiber. Biophysical Journal, 35:193–213.

Speakers

Marco Arieli Herrera Valdez

Lionardo Truqui

Hector Chaparro-Reza

Vania Victoria Villegas

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P111: A model of 3D eye movements reflecting spatial orientation estimation

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P111 A model of 3D eye movements reflecting spatial orientation estimation

Yusuke Shinji1,Yutaka Hirata*2,3,4

1. Dept. Computer Science, Chubu Univ. Graduate School of Engineering, Kasugai, Japan
2. Dept. AI and Robotics, Chubu Univ. College of Science and Engineering, Kasugai, Japan
3. Center for mathematical science and artificial intelligence, Chubu Univ., Kasugai, Japan
4. Chubu University Academy of Emerging Sciences, Chubu Univ., Kasugai, Japan

*Email: yutaka@isc.chubu.ac.jp

Introduction
Spatial orientation (SO) refers to the estimated self-motion state, formed by integrating multiple sensory inputs. Accurate SO formation is crucial for animals to navigate safely through their environment. However, errors in SO estimation can occur, leading to spatial disorientation (SDO).A typical example of SDO is the somatogravic illusion, in which a forward linear acceleration is erroneously perceived as an upward tilt. The vestibulo-ocular reflex (VOR) is driven by the estimated head motion, generating counter-rotations of the eyes to stabilize vision. Thus, the VOR is a reflection of SO, particularly of head motion states. Currently, we developed a 3D VOR model to elucidate the neural algorithms underlying SO formation.

Methods
The model was configured within the Kalman filter (KF) framework, which estimates hidden physical states from noisy sensor data. In this framework, active 3D head motion is the input to the sensors while passive head motion and 3D visual motion are treated as process noise. These motions are detected by the otolith, semicircular canals, and retina whose outputs are transmitted to the brain. The KF incorporates corresponding sensor models that generate sensory predictions, which are compared with actual sensory outputs. The resulting sensory prediction errors are used to update the estimated head motion state through the KF algorithm.The VOR eye velocity is then produced in the opposite direction to the 3D head motion estimate.

Results
To evaluate the model, we first simulated the somatogravic illusion in goldfish that we recently discovered [1]. The model successfully reproduced the goldfish 3D VOR reflecting somatogravic illusion.Specifically, in the KF, lateral linear head acceleration was misestimated as head roll tilt, resulting in a vertical VOR in goldfish, while head roll tilt motion was correctly estimated as roll tilt.Next, we simulated two representative human vestibular illusions: Off-vertical axis rotation at a constant angular velocity, and Post-rotatory tilt after earth-vertical axis rotation. In both cases, the model misestimated linear head acceleration, reproducing known perceptual errors in humans.

Discussion
These results suggest that our 3D VOR KF model effectively captures the neural computational mechanisms underlying SO formation from noisy sensory signals.Previous studies have demonstrated the cerebellar nodulus and uvula play a critical role in SO formation, specifically in distinguishing head tilt against gravity from head linear translational acceleration [2].As a next step we investigate how the well-characterized cerebellar neuronal circuitry and its synaptic learning rules implement the KF algorithm, utilizing our artificial cerebellum [3]. Understanding this relationship will provide insights into how the brain optimally estimates self-motion and resolves sensory ambiguities.

Acknowledgements
Supported by JST CREST (Grant Number: JPMJCR22P5) and JSPS KAKENHI (Grant Number: 24H02338)

References
1. Tadokoro, S., Shinji, Y., Yamanaka, T., Hirata, Y. (2024). Learning capabilities to resolve tilt-translation ambiguity in goldfish.Front Neurol, 15:1304496. https://10.3389/fneur.2024.1304496
2. Laurens, J. (2022). The otolith vermis: A systems neuroscience theory of the Nodulus and Uvula.Front Neurosci, 16:886284. https://10.3389/fnsys.2022.886284
3. Shinji, Y., Okuno, T., Hirata, Y. (2024). Artificial cerebellum on FPGA: realistic real-time cerebellar spiking neural network model capable of real-world adaptive motor control.Front Neurosci, 18:1220908. https://10.3389/fnins.2024.1220908

Speakers

Yutaka Hirata

Yusuke Shinji

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P112: Dopamine effects on firing patterns of in vitro hippocampal neurons

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P112 Dopamine effects on firing patterns of in vitro hippocampal neurons

Huu Hoang*1, Aurelio Cortese2

1Neural Information Analysis Laboratories, ATR Institute International, Kyoto, Japan
2Computational Neuroscience Laboratories, ATR Institute International, Kyoto, Japan

*Email: hoang@atr.jp

Introduction:Dopamine plays a pivotal role in shaping hippocampal neural activity, yet its impact on network dynamics is not fully understood. We explored this by recording rat embryonic hippocampal neurons in vitro, using electrophysiological techniques, pharmacological manipulations, and spike train analysis. Our findings reveal that dopamine reduces network synchrony, broadening the range of burst dynamics—an effect absent with dopamine antagonists. This study deepens our insight into how dopamine signaling shapes functional hippocampal networks.

Methods:We cultured eight rat embryonic hippocampus samples and used MaxOne microelectrode arrays to record spiking activity from hundreds of electrodes. Baseline spikes were captured without dopamine, then dopamine was added gradually, and spikes were recorded. We assessed synchrony strength via spike coherence in 1 msec bins and examined dopamine’s effect on spike dynamics. Spike bursts (typically in 200-300 ms) were detected, and their similarity index was measured. Using affinity propagation on the similarity index, we identified repeating burst motifs, revealing insights into burst dynamics. We used linear mixed-effect models to statistically evaluate the influence of dopamine on the metrics of interest.

Results:Our study revealed that dopamine lowers synchrony strength, enhances network modularity, and restricts connectivity within modules, while broadening burst pattern variety. At higher dopamine concentrations (300-1000 μM), burst frequency rose, yet burst similarity dropped, with repeating motifs surging 40-50% above baseline. The reduction in synchrony caused by dopamine directly lessened burst pattern similarity, shown by a robust positive correlation between synchrony and similarity changes in eight samples. This relationship disappeared in samples treated with dopamine antagonists, underscoring dopamine’s critical influence on reorganizing network dynamics and its possible role in cognitive processes.

Discussion:We investigated dopamine’s impact on cultured hippocampal neurons using high-density electrode arrays, observing a rise in burst events with pronounced synchrony across hundreds of electrodes after incrementally adding dopamine, consistent with previous studies. This setup provided detailed network-level insights with cellular and millisecond precision, showing dopamine reduced spike synchrony while increasing the number of network modules with more restricted connectivity. Such reorganization may optimize information flow for cognitive functions like memory and decision-making. Dopamine also diversified burst patterns, boosting repeating motifs and lowering burst similarity—an effect blocked by antagonists. These findings suggest dopamine enhances distinct encoding in hippocampal circuits, offering potential implications for understanding cognition and schizophrenia therapies.

Acknowledgements
This study was supported by JST ERATO (JPMJER1801, "Brain-AI hybrid").

References
Hoang H, Matsumoto N, Miyano M, Ikegaya Y, Cortese A. (2025). Dopamine-induced relaxation of spike synchrony diversifies burst patterns in cultured hippocampal networks.Neural Networks; 181:106888. doi: 10.1016/j.neunet.2024.106888.

Speakers

Huu Hoang

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P113: Baseline firing rate of dopaminergic neurons modulates the dopamine response to stimuli

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P113 Baseline firing rate of dopaminergic neurons modulates the dopamine response to stimuli

M. Duc Hoang*1, Andrea R. Hamilton2,Timothy J. Lewis1, Stephen L. Cowen3,4and M. Leandro Heien2

1Department of Mathematics, University of California, Davis, Davis, CA, USA
2Department of Chemistry & Biochemistry, University of Arizona, Tucson, AZ, USA
3Department of Psychology, University of Arizona, Tucson, AZ, USA
4Evelyn F. McKnight Brain Institute, University of Arizona, Tucson, AZ, USA

*Email: mdhoang@ucdavis.edu

Introduction

Electrical brain stimulation (EBS) targeting dopaminergic (DA) neurons is a valuable tool for investigating dopamine circuits and treating diseases such as Parkinson's disease [1]. However, our understanding of how the temporal structure of stimuli interacts with the firing dynamics of DA neurons to regulate dopamine release remains limited. In this study, we experimentally measure changes in dopamine concentration in response to stimulation of the medial forebrain bundle and develop a data-driven mathematical model to describe this stimulus-evoked dopamine response. Our results demonstrate that the baseline firing rate (BFR) of DA neurons prior to electrical stimulation can strongly modulate the DA response.

Methods
In this study, we use fast-scan cyclic voltammetry (FSCV) to measure changes in dopamine (DA) concentration in the nucleus accumbens (NAc) in response to stimulation of the medial forebrain bundle (MFB) in anesthetized rats [2]. We then implement a modification of the Montague et al. model [3] of DA response to electrical stimulation of DA axons in the MFB. The model includes synaptic facilitation and depression, as well as feedback inhibition through D2 autoreceptors (D2AR). We fit model parameters to the FSCV data from multiple stimulation patterns simultaneously. Importantly, we account for the unknown baseline DA levels in our parameter fits. We also validate the model with additional experimental data sets.
Results
We observe a high degree of variability in the dopamine response in the NAc when the MFB is subjected to identical 20 Hz stimuli across several trials. Specifically, the peak change in DA concentration differs by ~40% between trials (Fig. 1A). Simulations of our model show a similarly large variation in peak DA concentration in response to 20 Hz stimulation when the baseline firing rate (BFR) of simulated DA neurons is varied from 0 to 6 Hz, even though the corresponding variation in baseline DA concentration is below 0.02M (Fig. 1B). We use phase-plane analysis to elucidate the mechanism underlying this phenomenon and describe how the phenomenon is influenced by BFR, dopamine reuptake, and D2AR inhibition.
Discussion
Our experimental and modeling results suggest that small fluctuations in baseline DA concentrations in the NAc due to changes in BFR of DA neurons, D2AR levels, or DA reuptake rates can significantly alter the DA response to MFB stimulation. Analysis of the model reveals that the underlying mechanism of this phenomenon involves the interplay of the firing rate of DA neurons, DA reuptake dynamics, and synaptic depression. These findings underscore the importance of BFR in modulating dopamine release during EBS, suggesting that BFR may influence the efficacy of EBS in treating disorders such as Parkinson’s disease, depression, and schizophrenia. This insight could inform the optimization of EBS protocols for therapeutic applications.

Figure 1. The baseline firing rate (BFR) leads to substantial variation in DA response to identical stimulation. A: The change in DA in the NAc in response to a 20 Hz periodic stimulation applied to the MFB. 3 trials are shown of the same stimulus. B: DA concentration profiles predicted by the modified Montague et al. model in response to 20 Hz stimulation for 5 different BFRs (0-6 Hz).
Acknowledgements
Funding for this project is provided by the National Institutes of Health (R01 NS123424-01). A.R.H. was funded by T32 GM008804.
References
[1] https://doi.org/10.3171/2019.4.JNS181761
[2] https://doi.org/10.1021/acschemneuro.4c00115
[3] https://doi.org/10.1523/JNEUROSCI.4279-03.2004

Speakers

Minh Duc Hoang

Andrea Raye Hamilton

Timothy James Lewis

Stephen Leigh Cowen

Michael Leandro Heien

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

17:20 CEST

P141: Intrinsic neuronal properties shape local circuit inhibition in primate prefrontal cortex

Sunday July 6, 2025 17:20 - 19:20 CEST

Passi Perduti

P141 Intrinsic neuronal properties shape local circuit inhibition in primate prefrontal cortex

Nils A. Koch*1, Benjamin W. Corrigan2,3,4, Julio C. Martinez-Trujillo3,4,5, Anmar Khadra1,6

1Integrated Program in Neuroscience, McGill University, Montreal, QC, Canada
2Department of Biology, York University, Toronto, ON, Canada
3Department of Clinical Neurological Sciences, London Health Sciences Centre, Western University, London, ON, Canada
4Department of Physiology and Pharmacology, Schulich School of Medicine and Dentistry, Western University, London, ON, Canada
5Western Institute for Neuroscience, Western University, London, ON, Canada
6Department of Physiology, McGill University, Montreal, QC, Canada

*Email: nils.koch@mail.mcgill.ca
Introduction

Intrinsic neuronal properties play a key role in neuronal circuit dynamics. One such property evident during step-current stimulation is intrinsic spike frequency adaptation (I-SFA), a feature noted to be important for in vivo activity [1] and computational capabilities of neurons [2, 3, 4, 5, 6, 7, 8, 9, 10, 11]. In behaving animals, extracellular recordings exhibit extrinsic spike frequency adaptation (E-SFA) in response to sustained visual stimulation. However, the relationship between the I-SFA measured in vitro, typically in response to constant step-current pulses, and the E-SFA described in vivo during behavioral tasks, in which the inputs into a neuron are likely variable and difficult to measure, is not well characterized.
Methods
To investigate how I-SFA in neurons isolated from brain networks contributes to E-SFA during behavior, we recorded responses of macaque lateral prefrontal cortex neurons in vivo during a visually guided saccade task and in acute brain slices in vitro. Units recorded in vivo and neurons recorded in vitro were classified as broad spiking (BS) putative pyramidal cells and narrow spiking (NS) putative inhibitory interneurons based on spike width. To elucidate how in vitro I-SFA contributes to in vivo E-SFA, we bridge the gap between the in vitro and in vitro recordings with a data-driven hybrid circuit model in which NS neurons fit to the in vitro firing behavior are driven by local BS input.
Results
Both BS and NS units exhibited E-SFA in vivo. In acute brain slices, both cell types displayed differing magnitudes of I-SFA but with timescales similar to E-SFA. The model NS cell responses show longer SFA than observed in vivo. However, introduction of inhibition of NS cells to the model circuit removed this discrepancy and reproduced the in vivo E-SFA, suggesting a vital role of local circuitry in dictating task-related in vivo activity. By exploring the relationship between individual neuron I-SFA and hybrid circuit model E-SFA, the contribution of I-SFA to E-SFA is uncovered. Specifically, this contribution is dependent on the timescale of I-SFA and modulates in vivo response magnitudes as well as E-SFA timescales.
Discussion
Our results indicate that both I-SFA and inhibitory circuit dynamics contribute to E-SFA in LPFC neurons during a visual task and highlight the contribution of both single neurons and network dependent computations to neural activity underlying behavior. Furthermore, the interaction between excitatory input and I-SFA demonstrates that inhibitory cortical neurons do not solely contribute to the local circuit inhibition by altering the sign of signals (i.e. from excitation to inhibition) and that the intrinsic properties of NS neurons contribute to their activity in vivo. Consequently, large models of cortical networks as well as artificial neuronal nets that emphasize network connectivity may benefit from including intrinsic neuronal properties.

Acknowledgements
This work was supported by a Natural Sciences and Engineering Research Council of Canada (NSERC) Discovery Grant to A.K.; Canadian Institutes of Health Research (CIHR), NSERC, Neuronex (ref. FL6GV84CKN57) and BrainsCAN grants to J.C.M.-T.; and an NSERC Postgraduate Scholarship-Doctoral Fellowship to N.A.K..
References
1. https://doi.org/10.3934/mbe.2016002
2. https://doi.org/10.1016/j.cub.2020.11.054
3. https://doi.org/10.1007/s10827-007-0044-8
4. https://doi.org/10.1523/JNEUROSCI.4795-04.2005
5. https://doi.org/10.1038/s41467-017-02453-9
6. https://doi.org/10.1523/ENEURO.0305-18.2020
7. https://doi.org/10.1016/j.conb.2013.11.012
8. https://doi.org/10.1016/j.biosystems.2022.104802
9. https://doi.org/10.1007/s00422-009-0304-y
10. https://doi.org/10.1523/JNEUROSCI.1792-08.2008
11. https://doi.org/10.1016/j.neuron.2016.09.046

Speakers

Nils Koch

Benjamin W. Corrigan

Julio Martinez-Trujillo**

Anmar Khadra

Sunday July 6, 2025 17:20 - 19:20 CEST
Passi Perduti

Posters, Poster session 1

08:30 CEST

Registration

Monday July 7, 2025 08:30 - 19:00 CEST

Registration/organization

09:00 CEST

Announcements and Keynote #3: Ken Miller

Monday July 7, 2025 09:00 - 10:10 CEST

Auditorium - Plenary Room

Speakers

Ken Miller

Monday July 7, 2025 09:00 - 10:10 CEST
Auditorium - Plenary Room

Keynote

10:10 CEST

Coffee break

Monday July 7, 2025 10:10 - 10:40 CEST

Break

10:40 CEST

Oral session 3: Perturbing the brain

Monday July 7, 2025 10:40 - 12:30 CEST

Auditorium - Plenary Room

Monday July 7, 2025 12:10 - 12:30 CEST
Auditorium - Plenary Room

Oral session talk, Oral session 3

12:30 CEST

Lunch break

Monday July 7, 2025 12:30 - 14:00 CEST

Break

12:30 CEST

OCNS Board meeting

Monday July 7, 2025 12:30 - 14:00 CEST

TBA

Monday July 7, 2025 12:30 - 14:00 CEST
TBA

Registration/organization

14:00 CEST

Oral session 4: Modeling Disease

Monday July 7, 2025 14:00 - 15:50 CEST

Auditorium - Plenary Room

Monday July 7, 2025 15:30 - 15:50 CEST
Auditorium - Plenary Room

Oral session talk, Oral session 4

15:50 CEST

Coffee break

Monday July 7, 2025 15:50 - 16:20 CEST

Break

16:20 CEST

Poster session 2

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Poster session

16:20 CEST

P001: A Recurrent Neural Network Model of Cognitive Map Development

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P001 A Recurrent Neural Network Model of Cognitive Map Development

Marco P. Abrate*1, Tom J. Wills1, Caswell Barry1

1Department of Cell and Developmental Biology, University College London, London, UK

*Email: marco.abrate@ucl.ac.uk
Introduction
Animals use an allocentric cognitive map of self-location, constructed from sequential egocentric observations, to navigate flexibly[1-4]. In the hippocampal formation, spatially modulated neurons support navigation, such as place cells, Head Direction (HD) cells, grid cells, and boundary cells[5-8]. The early development of these neurons is well characterised[9] but the mechanisms driving maturation and the relative timing of their emergence are unclear. We hypothesize that changes in locomotion shape the development of spatial representations. Combining behavioural analysis with a Recurrent Neural Network (RNN), we prove that movement statistics determine the development of spatial tuning, mirroring biological timelines.
Methods
Rats from post-natal day 12 (P12) to P25[10-12] were grouped according to their movement statistics. Rodent trajectories were simulated, using the RatInABox toolbox[13], in a square arena matching these locomotion stages. An RNN was trained to predict upcoming visual stimuli based on previous visual and vestibular inputs - mimicking the predictive coding function of biological systems[14]. The hidden units' activity was analysed against the position and the facing direction of the agent. Finally, these units were classified as place units based on their spatial information content or as HD units based on their Rayleigh vector length and KL divergence vs a uniform distribution - standard metrics for hippocampal neural recordings.
Results
Behavioural analysis revealed three distinct stages of locomotion during development with median ages P14, P15, and P21, respectively (Fig. 1a). The RNN trained on adult-like locomotion (Fig. 1b), solving the predictive task with biologically plausible inputs, showed spatially tuned units resembling hippocampal place and head direction cells (Fig. 1c). Crucially, when trained separately on simulated locomotion styles corresponding to the identified developmental stages, the model recapitulated the progressive emergence of spatial tuning observed experimentally. Specifically, spatial measures and consequently the number of units classified as place and head direction neurons steadily increased with improved locomotion (Fig. 1d).
Discussion
Our model establishes locomotion-dependent sensory sampling as a sufficient mechanism for cognitive map formation, extending predictive coding theories[3,4,15]. The RNN's ability to replicate spatial cell maturation patterns suggests that sensory-motor experience significantly shapes hippocampal spatial tuning. Furthermore, our results inform how manipulations of locomotion or sensory inputs could influence the development of spatial representations, which can then be tested in real-world experiments. Future work will directly compare the RNN's units with hippocampal neurons through Representational Similarity Analysis, search what drives grid patterns formation in our model, and investigate the changes in the geometry of the latent space.
Figure 1. (a) 3-d UMAP representation of rats’ movement statistics coloured into locomotion stages. (b) Example of an agent’s trajectory (left) and snapshot of current visual input (right). (c) Architecture of the RNN. The latent space’s units are analysed for spatial responses. (d) Trend in the number of the RNN’s units classified as Place units (left), HD units (right), or Place and HD units (both).
Acknowledgements
NA
References

1. doi.org/10.1017/S0140525X00063949
2. doi.org/10.1037/h0061626
3. doi.org/10.1016/j.tics.2018.07.006
4. doi.org/10.1038/nn.4650
5. doi.org/10.1016/0006-8993(71)90358-1
6. doi.org/10.1523/JNEUROSCI.10-02-00420.1990
7. doi.org/10.1038/nature03721
8. doi.org/10.1523/JNEUROSCI.1319-09.2009
9. doi.org/10.1002/wcs.1424
10. doi.org/10.1126/science.1188224
11. doi.org/10.1016/j.neuron.2015.05.011
12. doi.org/10.1016/j.cub.2019.01.005
13. doi.org/10.7554/eLife.85274
14. doi.org/10.1017/S0140525X12000477
15. doi.org/10.1016/j.cell.2020.10.024
@font-face {font-family:"Cambria Math"; panose-1:2 4 5 3 5 4 6 3 2 4; mso-font-charset:0; mso-generic-font-family:roman; mso-font-pitch:variable; mso-font-signature:-536870145 1107305727 0 0 415 0;}@font-face {font-family:Aptos; panose-1:2 11 0 4 2 2 2 2 2 4; mso-font-charset:0; mso-generic-font-family:swiss; mso-font-pitch:variable; mso-font-signature:536871559 3 0 0 415 0;}p.MsoNormal, li.MsoNormal, div.MsoNormal {mso-style-unhide:no; mso-style-qformat:yes; mso-style-parent:""; margin:0cm; mso-pagination:widow-orphan; font-size:12.0pt; font-family:"Aptos",sans-serif; mso-ascii-font-family:Aptos; mso-ascii-theme-font:minor-latin; mso-fareast-font-family:Aptos; mso-fareast-theme-font:minor-latin; mso-hansi-font-family:Aptos; mso-hansi-theme-font:minor-latin; mso-bidi-font-family:"Times New Roman"; mso-bidi-theme-font:minor-bidi; mso-font-kerning:1.0pt; mso-ligatures:standardcontextual; mso-fareast-language:EN-US;}.MsoChpDefault {mso-style-type:export-only; mso-default-props:yes; font-family:"Aptos",sans-serif; mso-ascii-font-family:Aptos; mso-ascii-theme-font:minor-latin; mso-fareast-font-family:Aptos; mso-fareast-theme-font:minor-latin; mso-hansi-font-family:Aptos; mso-hansi-theme-font:minor-latin; mso-bidi-font-family:"Times New Roman"; mso-bidi-theme-font:minor-bidi; mso-fareast-language:EN-US;}div.WordSection1 {page:WordSection1;}

Speakers

Thomas Wills

Marco Pietro Abrate

Caswell Barry

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P108: Bayesian Inference Across Brain Scales

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P108 Bayesian Inference Across Brain Scales

M. Hashemi*1, N Baldy1, A. Ziaeemehr1, A. Esmaeili1, S. Petkoski1, M. Woodman1, V. Jirsa*1

1Aix Marseille Univ, INSERM, INS, Inst Neurosci Syst, Marseille, France
Email :Meysam.hashemi@univ-amu.fr/viktor.jirsa@univ-amu.fr

Introduction

The process of inference across spatiotemporalscalesis essential toidentifythe underlyingcausalmechanisms of brain computation and (dys)function. However, there remains a critical need for automated model inversion tools to estimate control (bifurcation) parameters from recordings acrossbrainscales, ideally including uncertainty.

Methods
In this work, we attempt to bridge this gap by providing efficient and automatic Bayesian inference operating across scales. We usethestate-of-the-art probabilistic machine learning tools employing likelihood-based (MCMC sampling [1, 2]) and likelihood-free (a.k.a. simulation-based inference [3, 4]) approaches.

Results
We demonstrate inference on the parameters and dynamics of spiking neurons, their mean-field approximation at the regional level, and brain network models. We show the benefits of incorporatingprior andinference diagnostics, leveraging self-tuning Monte Carlo strategies for unbiased sampling, and deep density estimators for efficient transformations[5]. The performance of these methods is then demonstratedfor causal inference inepilepsy [6], multiple sclerosis [7], focal intervention [8], healthy aging [9], and social facilitation [10].

Discussion
This work shows potential to improve hypothesis evaluation in across brain scales through uncertainty quantification, and contribute to advances in precision medicine by enhancing the predictive power of brain models.
Figure 1. Bayesian inference across brain scales. (A) Based on Bayes’ theorem, background knowledge about control parameters (expressed as a prior distribution), is combined with information from observed data (in the form of a likelihood function) to determine the posterior distribution. (B) Examples of the observed and predicted data features.
Acknowledgements
This research has received funding from EU’s Horizon 2020 Framework Programme for Research and Innovation under the Specific Grant Agreements No. 101147319 (EBRAINS 2.0 Project), No. 101137289 (Virtual Brain Twin Project), and government grant managed by the Agence Nationale de la Recherche reference ANR-22-PESN-0012 (France 2030 program).
References

[1] DOIhttps://doi.org/10.1016/j.neuroimage.2020.116839
[2]DOI:10.1162/neco_a_01701
[3]DOI:10.1088/2632-2153/ad6230
[4]Doi:https://doi.org/10.1101/2025.01.21.633922
[5]DOI:https://doi.org/10.1101/2024.10.25.620245
[6]DOI:https://doi.org/10.1016/j.neunet.2023.03.040
[7]DOI:10.1016/j.isci.2024.110101
[8]DOI:https://doi.org/10.1101/2023.09.08.556815
[9]DOI:https://doi.org/10.1016/j.neuroimage.2023.120403
[10]DOI:https://doi.org/10.1101/2024.09.09.612006

Speakers

Marmaduke Woodman

Spase Petkoski

Viktor Jirsa

Nina Baldy

Meysam Hashemi

Abolfazl Ziaeemehr

Amirhossein Esmaeili

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P114: Modelling the bistable cortical dynamics of the sleep-onset period

P118: Simulating interictal epileptiform discharges in a whole-brain neural mass model with calcium-mediated bidirectional plasticity

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P118 Simulating interictal epileptiform discharges in a whole-brain neural mass model with calcium-mediated bidirectional plasticity

MehmetAlihanKayabas1, Elif Köksal-Ersöz2,3,Linda-Iris JosephTomy1,Pascal Benquet1, Isabelle Merlet1, Fabrice Wendling1
¹UnivRennes, INSERM, LTSI – UMR 1099, Rennes F-35000, France
²InriaLyonResearchCentre, Villeurbanne 69603, France
³CophyTeam, Lyon NeuroscienceResearchCenter, INSERM UMRS 1028, CNRS UMR 5292, Université Claude Bernard Lyon 1, Bron 69500, France
*Email:malihankayabas@gmail.com
Introduction

Whole-brainmodeling of interictal epileptiform dischargesoffers a promising approach to optimize transcranial direct current stimulation (tDCS) protocols, byidentifyingspecific regions of the brain involved in the epileptic network [1,2]. In this study, we investigated the synaptic plasticity induced bytDCSin a whole-brain network of connected neural mass models (NMMs) [3, 4], which we extended by implementing the calcium-mediated synaptic plasticity mechanisms based on our recent study [5]. We studied the impact oftwolocal parameters:synapticdepression(θd)andpotentiation(θp)thresholds, on long-term depression (LTD) and long-term potentiation (LTP) undertDCS.

Methods
The activity of each node of the network was simulated by NMMs including excitatory and inhibitory neuronal subpopulations. The nodes are interconnected by a structural connectivity matrix fromthe HumanConnectome Project [6]. We tuned parameters of NMMstosimulateinterictal epileptiformdischarges (IEDs), alpha-band activity (8-12 Hz), and background activity. We assumed that the electrical stimulation affects the mean membrane potential of excitatory neuronal subpopulations. We varied thedepression andpotentiation threshold parametersin different subnetworks and simulated the system for 15 min for each condition.Two metrics were evaluated:Functional connectivity calculated using a non-linear correlation coefficient, and mean amplitude per channel.
Results
Under most conditions both the signal amplitude and node strength decreased.(Fig. 1).Except forall_nodes_θdconditionwherethedepression thresholdwas increasedacross all nodeswhichreduced LTD activity, resulting in an increase in strength of epileptic nodes.Regionally,the parietal nodesshowedthemost significantreductionswhile the frontal nodesthe least significantvariations. An increase in potentiation threshold across all nodes (all_nodes_θpcondition) resulted inthehighestreduction in bothamplitude and strength.,When bothθdandθpwere increased simultaneously,the decrease in strength of epileptic nodes was even more pronounced, while increasingθpalone in the occipital nodes did not yield a reduction in epileptic node strength.
Discussion
Variation insynapticplasticity thresholds alters whole-brain network dynamics. In nodesexhibitingalpha-band activity, decreased node strength lowers signal amplitude without changing frequency. Inepileptogenic nodes, reduced node strength leads to lower IED frequency anddesynchronization between two regions of the epileptogenic zone, while increased strength has the opposite effect.As of this day, there is no consensus in the literature on the effect oftDCSon alpha-band[7,8] or on the IED frequency [9, 10].In future studies, we willstudyour model further to elucidate the mechanisms and role oftDCStreatmentin focal epilepsy.

Figure 1. (A) percentage difference in node strength relative to basal level. (B) percentage difference in amplitude relative to basal level. (C) Examples of LFP signals for left lateral occipital (alpha) and precentral (epileptic) nodes before (blue) and after (red) the increase in potentiation threshold. θp: Potentiation threshold and θd: Depotentiation threshold.* Denotes p-value < 0.05 Kruskal-Wallis
Acknowledgements
This project has received funding from the European Research Council under the European Union’s Horizon 2020 research and innovation program (No 855109).
References
1.https://doi.org/10.1093/med/9780199746545.003.0017
2.https://doi.org/10.1093/brain/awz269
3.https://doi.org/10.1016/j.softx.2024.101924
4. https://doi.org/10.1088/1741-2552/ac8fb4
5.https://doi.org/10.1371/journal.pcbi.1012666
6.https://doi.org/10.1016/j.neuroimage.2021.118543
7.https://doi.org/10.3389/fnhum.2013.00529
8.https://doi.org/10.1038/s41598-020-75861-5
9.https://10.1016/j.brs.2016.12.005
10.https://doi.org/10.1016/j.eplepsyres.2024.107320

Speakers

Mehmet Alihan Kayabas

Elif Koksal Ersoz

Pascal Benquet

Linda Iris Joseph Tomy

Isabelle Merlet Paquin

Fabrice Wendling

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P119: Computational modeling of the cumulative neuroplastic effects of repeated direct current stimulation

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P119 Computational modeling of the cumulative neuroplastic effects of repeated direct current stimulation

Linda-Iris JosephTomy1, Elif Köksal-Ersöz2,3,MehmetAlihanKayabas1, Pascal Benquet1, Fabrice Wendling1
¹UnivRennes, INSERM, LTSI – UMR 1099, Rennes F-35000, France
²InriaLyonResearchCentre, Villeurbanne 69603, France
³CophyTeam, Lyon NeuroscienceResearchCenter, INSERM UMRS 1028, CNRS UMR 5292, Université Claude Bernard Lyon 1, Bron 69500, France
*Email:malihankayabas@gmail.com
Introduction

Metaplasticitymodulates theneuroplasticabilityofneurons/synapsesin orderto maintain it withinafunctionalphysiological range.In conditionssuch asepilepsy, where neuroplasticitymayevolve pathologically, this metaplastic property of synapseswould also bedisrupted[1].Transcranial direct current stimulation (tDCS)isa non-invasivetechniquethatcan modulateneuroplasticity.Repeatedsessions oftDCScanimprove the likelihood of inducingseizure reduction in patients with refractory epilepsy[2].TheeffectoftDCSon neuroplasticityhas also been shown todependon the ongoingneuronalactivity andneuroplastic propertiesofthe stimulated brain regions.

Methods
Computational modeling[3]was usedtoidentify‘functional’ and‘dysfunctional’ metaplastic conditions.The modelconsistedof an epileptogeniczone(EZ)connectedto anirritatedzone (IZ).We assumedthe potentiation threshold(ϴp)discriminatedbetween the‘functional’and‘dysfunctional’metaplastic conditions.We evaluated the variation in connectivity strengthbyinitiatingthe modelfrom depressed and potentiated states for the metaplastic conditionsfordifferent frequencies of interictal activity from the EZ.Theeffectof repeatedtDCSwasinvestigated.Variations in connectivity strengthfor different frequencies ofongoing neuronalactivitywere assessedby plottingthe frequency response function (FRF).
Results
In the‘functional’metaplastic condition, the connectivity strength from EZ to IZwasprevented from being potentiated orevolved towards depression.Whereas, in the ‘dysfunctional’ metaplastic condition, theconnectivity strength tendedto evolve towards potentiation.Further,a decrease in ϴpled totheexpansion of epileptic activity in this network.Under repetitivetDCSapplication,weobserveda downwardshift inthe FRF, suggesting that repetitivetDCScould promotelong term depression.
Discussion
In thisstudy, we exploredhow functional and dysfunctionalmetaplasticconditions affect neuroplasticityin an epileptic network.Theimpact of varying ϴpto switch between thesemetaplasticconditionsreflectedthe relationship betweenmetaplasticityandepileptogenicity, as also seen in animal studies [1].Based on the variationsin the FRF observed here, it may be possible to designtDCSprotocolsto depressthe connectivityfromthe EZ to other IZs. Thismay thenimprovestimulationoutcomes.

Acknowledgements
This project has received funding from the European Research Council under the European Union’s Horizon 2020 research and innovation program (No 855109).
References
● https://doi.org/10.1371/journal.pcbi.1012666

● https://doi.org/10.1155/2017/8087401
● https://doi.org/10.1016/j.brs.2019.09.006

Speakers

Mehmet Alihan Kayabas

Elif Koksal Ersoz

Pascal Benquet

Linda Iris Joseph Tomy

Fabrice Wendling

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P120: Critical neuronal avalanches emerge from excitation-inhibition balanced spontaneous activity

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P120 Critical neuronal avalanches emerge from excitation-inhibition balanced spontaneous activity

Maxime Janbon1, Mateo Amortegui1, Enrique Hansen1, Sarah Nourin1, Germán Sumbre1,Adrián Ponce-Alvarez*2,3,4

1Institut de Biologie de l’ENS (IBENS), Département de biologie, École normale supérieure, CNRS, INSERM, Université PSL, 75005 Paris, France
2Departament de Matemàtiques, Universitat Politècnica de Catalunya, 08028 Barcelona, Spain.
3Institut de Matemàtiques de la UPC - Barcelona Tech (IMTech), Barcelona, Spain.
4Centre de Recerca Matemàtica, Barcelona, Spain.

*Email : adrian.ponce@upc.edu

Introduction

Neural systems exhibit cascading activity patterns called neuronal avalanches that follow power-law statistics, a hallmark of critical systems. Theoretical models [1,2] and in vitro studies [3] suggest that the excitation-inhibition (E/I) balance is a key factor in the self-organization of criticality. However, how E and I dynamics evolve and interact during in vivo neuronal avalanches remains unclear.
Here, we investigated E and I neuron contributions to spontaneous neuronal avalanches using double-transgenic zebrafish expressing cell-type-specific fluorescence proteins and calcium indicators. Furthermore, we built a stochastic E-I network model to explore how critical avalanches depend on the E/I ratio.
Methods
We monitored spontaneous neuronal activity in the optic tectum of 10 zebrafish larvae using selective-plane illumination microscopy (SPIM). Double-transgenic larvae expressing GCaMP6f and mCherry under the Vglut2a promoter (glutamatergic) were combined with immunostaining for GABAergic and cholinergic neurons, allowing identification of excitatory (E), inhibitory (I), and cholinergic (Ch) neurons.
We modelled the collective activity of E and I neurons using a model of critical dynamics that combines stochastic Wilson-Cowan equations[1,4],spatial embedded neuronal connectivity,and a spike-to-fluorescence convolutional model. Critical avalanches arise throughbalanced amplification[1] at a phase transition.
Results
Our results show that spontaneous fluctuations in E and I activity influenced neuronal avalanche statistics in the zebrafish optic tectum. Neuronal avalanches approached criticality when excitatory and inhibitory activity were balanced. Notably, the model accurately captured the observed avalanche statistics and their sensitivity to E/I fluctuations around a critical point defined by balanced excitatory and inhibitory synaptic strengths. Furthermore, the model allowed us to evaluate the statistics of neuronal avalanches derived from different simulated signals, representing calcium events or spiking activity. For both signals, the model's critical exponents align with experimental findings from calcium imaging and electrophysiology [5].
Discussion
Extensive research underscores the functional benefits of E/I balance and critical dynamics. Balanced networks enhance signal amplification, response selectivity, noise reduction, stability, memory, and plasticity [6-8], while critical dynamics optimize information processing [9-11]. Here, we showed that neuronal avalanche statistics and their dependence on spontaneous E/I fluctuations in the zebrafish optic tectum align with a model reaching criticality for balanced E and I couplings. Our study provides a framework to dissect the relationship between criticality and E/I balance, by manipulating the E/I ratio in vivo. Future integration of optogenetics into the present experiments and model will further clarify this interplay.

Acknowledgements
This study was supported by the Project PID2022-137708NB-I00 funded by MICIU/AEI /10.13039/501100011033 and FEDER, UE.A. Ponce-Alvarezwas supported by a Ramón y Cajal fellowship (RYC2020-029117-I) funded by MICIU/AEI/10.13039/501100011033 and “ESF Investing in your future”. G. Sumbre was supported by ERC CoG 726280.
References
1.https://doi.org/10.1371/journal.pcbi.1000846
2.https://doi.org/10.1523/JNEUROSCI.5990-11.2012
3.https://doi.org/10.1523/JNEUROSCI.4637-10.2011
4.https://doi.org/10.1371/journal.pcbi.1008884
5.https://doi.org/10.1126/sciadv.adj9303
6.https://doi.org/10.1088/0954-898X_6_2_001
7.https://doi.org/10.1126/science.274.5293.1724
8.https://doi.org/10.1016/j.neuron.2011.09.027
9.https://doi.org/10.1177/1073858412445487
10.https://doi.org/0.1523/JNEUROSCI.3864-09.2009
11.https://doi.org/10.1016/j.neuron.2018.10.045

Speakers

Maxime Janbon

Mateo Amortegui

Enrique Hansen

Sarah Nourin

German Sumbre

Adrián Ponce-Alvarez

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P121: Effects of the nonlinearity, kinetics and size of the gap junction connections on the transient dynamics of coupled glial cells

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P121 Effects of the nonlinearity, kinetics and size of the gap junction connections on the transient dynamics of coupled glial cells

Predrag Janjic*1, Dimitar Solev1, Ljupco Kocarev 1

1Research Center for Computer Science and Information Technologies, Macedonian Academy of Sciences and Arts, Skopje, North Macedonia

*Email: predrag.a.janjic@gmail.com
Introduction- The complex structure of massive couplings among the glial cells is not fully neurobiologically resolved, preventing realistic quantitative models. Despite the ongoing ultrastructural studies and elucidating published research focusing on glia[1] we can't extract statistical data on the number of gap junction (GJ) connections and their size. The nonlinear dependence of GJ conductance on the transjunctional voltage Vj, slow kinetics, and the GJ size-effect[2] on junction polarization suggest a rich repertoire of transient dynamic instabilities of resting glia when invaded by spreading depolarizations. Known limitations of the glial electrophysiology in-situ to measure GJ-coupled cells warrant qualifying suitable models.

Methods- We introduce a detailed point model of a coupled astrocytic cell - including several currents in the membrane kinetics and the nonlinear coupling with inactivation kinetics. Using the paradigm of a single active site in 1-d array of coupled astrocytes the main focus was on describing the bifurcations of the resting voltage Vr in the inner cell. Timescale separation allowed simplifying assumptions that enable formulating an ODE model of a "self-coupled cell", SCC. For stability analysis of such a model the 2nd cell is connected to a depolarized immediate neighbor on one-side, and a still quiet cell on the other side, both represented as fixed voltages Vdr and Vr. The numerical simulations were done on connected 1-d array.
Results- We explored the stability of the SCC in case of altered steady-state I-V curve, displaying N-shaped nonlinearity and generically present saddle-node (S-N) structure[3]. Newly introduced RMP is markedly more depolarized. The separate N-shaped nonlinearity introduced by the coupling enriched the S-N structure in all parameter perturbations. Typical cases were appearances of (a) fold limit cycle window within the range of fold curve, accompanied by a noise-induced bistability switching, or (b) a stable limit-cycle in the moderate coupling strength range. Not all of the observed dynamical regimes in the SCC survive in numerical simulation of an 1-d array, but in all cases we observed traveling front for the corresponding parameters.
Discussion- Emerging evidence from voltage imaging suggests that astrocytes do not respond dynamically as a homogeneous compartment, displaying strong variations in their depolarization between the collateral processes, or when compared to the cell body. In case of altered I-V curves they are generically prone to multistability. We observed enriched multistability scenarios in the passive response of GJ-coupled astrocytes under very basic conditions. We believe it motivates adding additional level of biophysical detail to the GJ connections and the topology of glial networks. Such groundwork is needed to extend the glial models with the advanced dynamical features of neuromodulation of their glutamate and GABA transporters and receptors.

Acknowledgements
The authors are grateful for the experimental recordings from isolated astrocytes shared by Prof. Christian Steinheauser, form the Institute of Cellular Sciences (IZN), School of Medicine, University of Bonn, Germany. PJ and DS were partially funded by R01MH125030 from the National Institute of Mental Health in US.
References
1. Aten,S., et al,(2022) Ultrastructural view of astrocyte arborization, astrocyte-astrocyte and astrocyte-synapse contacts, intracellular vesicle-like structures, and mitochondrial network, Prog Neurobiol, (213), 102264.
2. Wilders,R., and Jongsma,H.J.,(1992) Limitations of the dual voltage clamp method in assaying conductance and kinetics of gap junction channels, Biophys J 63(4), 942-953.
3. Janjic,P., Solev,D., and Kocarev,L.,(2023) Non-trivial dynamics in a model of glial membrane voltage driven by open potassium pores, Biophys J 122(8), 1470-1490.

Speakers

Predrag Janjic

Dimitar Solev

Ljupco Kocarev

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P122: Encoding visual familiarity for navigation in a mushroom body SNN trained on ant-perspective views

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P122 Encoding visual familiarity for navigation in a mushroom body SNN trained on ant-perspective views

Oluwaseyi Oladipupo Jesusanmi1,2, Amany Azevedo Amin2, Paul Graham1,2, Thomas Nowotny2
1Sussex Neuroscience, University of Sussex, Brighton, United Kingdom
2Sussex AI, University of Sussex, Brighton, United Kingdom
*Email: o.jesusanmi@sussex.ac.uk

Introduction

Ants can learn long visually guided routes with limited neural resources, mediated by the mushroom body brain region acting as a familiarity detector[1,2]. In the mushroom body, low dimensional input from sensory regions is projected into a large population of neurons, producing sparse representations of input information. These representations are learned via an anti-Hebbian process, modulated through dopaminergic learning signals. In navigation, the mushroom bodies guide ants to seek similar views to those previously learned on a foraging route. Here, we further investigate the role of mushroom bodies in ants’ visual navigation with a spiking neural network (SNN) model and 1:1 virtual recreations of ant visual experiences.
Methods
We implemented the SNN model in GeNN[3]. It has 320 Visual Projection Neurons (VPNs), 20000 Kenyon Cells (KCs), one Inhibitory Feedback Neuron (IFN) and one Mushroom Body Output Neuron (MBON). We used Deeplabcut to track ant trajectories in behavioural experiments. We used phone camera input into Neural Radiance Field (NeRF) and photogrammetry software for environment reconstruction. We used Isaac Sim and NVIDIA Omniverse to recreate views along ants’ movement trajectories from the perspective of the ants. We trained the SNN and comparator models (perfect memory and infomax[4]) on these recreations. We modelled inference across all traversable areas of the environment to test each model’s ability to encode navigational information.
Results
We produced familiarity landscapes for our SNN mushroom body and comparator models, showing differences between how they encode off-route (unlearned) locations. The mushroom body model produced navigation accuracy comparable to the other models. We found that the mushroom body model activity was able to explain trajectory data in trials where ants reached the target location. We found some views resulting in high familiarity did not appear in the training set. These views have similar image statistics to images in the training set, even if the view is from a different place in the environment. We found that ant trajectory routes with higher rates of oscillation improved learning, “filling-in” more of the familiarity landscapes.
Discussion
How the mushroom body would respond across all locations in a traversable environment is not known and is normally not feasible to study. Neural recording in ants remains difficult, and there are limited methods to have an ant systematically experience an entire experimental arena. We addressed this issue via simulation of biologically plausible neural activity while having exact control of what the model sees. Visual navigation models have been compared with mushroom body models in terms of navigation accuracy, but the familiarity landscape produced by the varied models has not been compared. Our investigation provides insights into how encoding of familiarity differs and leads to accurate navigation between models.

Acknowledgements.
References
● https://doi.org/10.1016/J.CUB.2020.07.013
● https://doi.org/10.1016/J.CUB.2020.06.030
● https://doi.org/10.1038/srep18854
● https://doi.org/10.1162/isal_a_00645

Speakers

Thomas Nowotny

Oluwaseyi Jesusanmi

Amany Amin

Paul Graham

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P123: Innovative Strategies to Balance Speed and Accuracy in P300-ERP Detection for Enhanced Online Brain-Computer Interfaces

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P123 Innovative Strategies to Balance Speed and Accuracy in P300-ERP Detection for Enhanced Online Brain-Computer Interfaces

Javier Jiménez*1, Francisco B. Rodríguez1

1Grupo de Neurocomputación Biológica, Departamento de Ingeniería Informática, Escuela Politécnica Superior, Universidad Autónoma de Madrid, Madrid, Spain
*Email: javier.jimenez01@uam.es

Introduction

Brain-Computer Interfaces (BCIs) interpret signals the brain generates to control devices. These signals can be related to known Event-Related Potentials (ERPs) registered with neuroimaging methods such as electroencephalography [1]. However, ERP detection requires many trials due to its low signal-to-noise ratio [2]. This detection method leads to the well-known speed-accuracy trade-off [3], as each trial adds to the time required for evoking ERPs. We propose a methodology for analyzing this trade-off using two new measures to find the best number of trials for the required accuracy. Finally, these measures were assessed using a P300-ERP dataset [4] to explore their potential as additional early-stopping methods in future online BCI setups.
Methods
In the literature, the speed-accuracy trade-off is usually studied by the employment of BCI measures such as the Information-Transfer Rate (ITR) [5]. However, these measures combine speed and accuracy within a single measure hindering BCI’s evaluation of such speed and accuracy separately. Considering these two concepts may be of interest to BCI users who would be able to decide whether they prefer a fast or accurate BCI in different scenarios. This work introduces two measures called Gain and Conservation which consider the amount of saved time and preserved accuracy, respectively, against a baseline BCI employing a Bayesian Linear Discriminant Analysis (BLDA) classifier to detect P300s.
Results
The new measures were tested against Hoffmann et. al. dataset [4] employing a BLDA classifier to detect P300s in combination with a traditional fixed-stop strategy based upon stopping after a fixed number of trials to evaluate the speed and accuracy of BCIs. For this paradigm, the expected behaviour of these measures would be to follow the speed-accuracy trade-off i.e. faster BCIs would correspond with inaccurate BCIs and vice-versa. This is because faster BCIs employ fewer trials and therefore have access to less information leading to worse P300 detection performances. Such behaviour can be seen in (Fig. 1) where the speed and accuracy of a BCI are represented by the Gain and Conservation, respectively.
Discussion
The described framework proposes two measures capable of evaluating BCIs’ speed and accuracy separately in contrast with other measures such as the ITR [5]. With these new measures, designers and users are provided with a controllable way to optimize BCIs towards different goals prioritizing one measurement over the other under demand. Furthermore, employing these measures offers detailed insights into the behaviors of different BCIs and early-stopping strategies [3] among other applications. To conclude, these measurements can be tracked during the BCI operation, which represents a key future direction of this work:leveraging the speed-accuracy trade-off of BCIs online.

Figure 1. Figure 1: Pseudo-online evolution along different trials from Hoffmann et. al. [4] of normalized Gain and Conservation measures for a fixed-stop strategy compared against its ITR.
Acknowledgements
This work was supported by the Predoctoral Research Grants of the Universidad Autónoma de Madrid (FPI-UAM) and by PID2023-149669NB-I00 (MCIN/AEI and ERDF – “A way of making Europe”).
References
1. 10.1016/0013-4694(88)90149-6
2. 10.1016/j.neuroimage.2010.06.048
3. 10.1088/1741-2560/10/3/036025
4. 10.1016/j.jneumeth.2007.03.005
5. 10.1016/S1388-2457(02)00057-3

Speakers

Francisco B Rodriguez

Javier Jiménez

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P124: Computational Prediction and Empirical Validation of Enhanced LTP Effects with Gentle iTBS Protocols

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P124 Computational Prediction and Empirical Validation of Enhanced LTP Effects with Gentle iTBS Protocols

Kevin Kadak*1,2, Davide Momi1, Zheng Wang1, Sorenza P. Bastiaens1,2, Mohammad P. Oveisi1,3, Taha Morshedzadeh1,2, Minarose Ismail1,4, Jan Fousek5, and John D. Griffiths1,2,6
1Krembil Centre for Neuroinformatics, Centre for Addiction and Mental Health, Toronto.
2Institute of Medical Sciences, University of Toronto
3Institute of Biomaterials and Biomedical Engineering, University of Toronto
4Department of Physiology, University of Toronto
5Central European Institute of Technology, Czech Republic
6Department of Psychiatry, University of Toronto
*Email: kevin.kadak@mail.utoronto.ca
Introduction

TMS is an established neuromodulatory technique for inducing and assessing cortical excitability changes. Intermittent theta-burst stimulation (iTBS), mimicking endogenous neural activity, yields clinical efficacy comparable to traditional protocols but with significantly shorter treatment durations [1,2]. Despite widespread use for depression treatments, iTBS suffers from high inter-subject response variability. We developed a computational model integrating calcium-dependent plasticity within corticothalamic circuitry to predict optimal iTBS parameters, subsequently validating these predictions through empirical testing of motor-evoked potentials (MEPs) across novel and canonical protocols.

Methods
Our computational model simulated iTBS-induced plasticity effects following 600 pulses in corticothalamic circuitry by varying pulse-burst ratios and inter-burst frequency parameters. We then conducted a mixed-measure experimental paradigm testing standard (Protocol A) and four novel iTBS protocols (B-E; 2-5 pulse-burst, 3-7 Hz). MEPs were recorded pre-stimulation (PRE) and post-stimulation (POST1, POST2) to assess induced plasticity effects. Mixed-effects modelling was performed to analyze group-level effects and response rates.
Results
Our model predicted that gentler stimulation protocols characterized by lower pulse-burst ratios and targeted inter-burst frequencies would maximize long-term potentiation (LTP) effects while reducing response variance. Empirical results confirmed these predictions, with Protocol C (3 pulses/burst, 3 Hz) capturing the highest response rate (60% vs 47% for standard iTBS) and Protocol B (2 pulses/burst, 5 Hz) driving the strongest LTP effects among responders. Notably, protocols with frequencies aligned to participants' alpha subharmonics further modulated plasticity effects in Protocol B, while higher-frequency protocols (Protocol D, 7 Hz) initially induced LTD, which later inverted to LTP.
Discussion
Our findings demonstrate that gentler protocols outperform standard iTBS in driving consistent LTP effects, with efficacy further modulated by resonance between stimulation frequency and endogenous alpha subharmonics. This research highlights an important mechanistic basis for induced plasticity effects pertaining to protocol intensity whereby lower intensity protocols appear to better engage neuroplasticity mechanisms and mitigate metaplastic saturation. We provide a mechanistic framework and empirical validation for enhancing LTP protocols and improving clinical outcomes in TMS treatments for neuropsychiatric disorders.

Acknowledgements
N/A
References
●
doi: 10.1016/j.biopsych.2007.01.018
●
doi: 10.1016/S0140-6736(18)30295-2

Speakers

Kevin Kadak

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P125: Modeling the biophysics of computation in the outer plexiform layer of the mouse retina

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P125 Modeling the biophysics of computation in the outer plexiform layer of the mouse retina

Kyra L. Kadhim*1, 2, Ziwei Huang1, Michael Deistler2, 3, Jonas Beck1, 2, Jakob H. Macke1, 2, 3, Thomas Euler4, Philipp Berens1, 2

1Hertie Institute for AI in Brain Health, University of Tübingen,Tübingen,Germany
2Tübingen AI Center,University of Tübingen,Tübingen, Germany
3Machine Learning in Science, University of Tübingen,Tübingen, Germany
4Institute for Ophthalmic Research, University of Tübingen,Tübingen,Germany

*Email: kyra.kadhim@uni-tuebingen.de
Introduction

The outer retina is a complex system of neurons that processes visual information, and it is experimentally accessible for the collection of multimodal data. What makes this system complex and nonlinear are mechanisms such as the phototransduction cascade, specialized ion channels, ephaptic feedback mechanisms, and the ribbon synapse [1]. These mechanisms are not typically included in network models that either fit neural data or perform tasks. In particular, optimizing the parameters of such models is computationally challenging with current modelling approaches, which do not include gradient-based optimization methods. However, ignoring such mechanisms limits the ability to capture the computations performed by the retina.
Methods
We developed a fully-differentiable, biophysically-detailed model of the outer plexiform layer of the mouse retina and optimized its parameters with gradient descent. We implemented our model using the new software library Jaxley [2] which inherits functionality from the state of the art machine learning library JAX. In our model, we have so far implemented the phototransduction cascade [3], ion channels [4], and ribbon synapse [5], and we fit their parameters to electrophysiology and neurotransmitter release data. We then optimized the synaptic conductances of the model to classify images with different contrast levels and global luminance levels and analyzed the trained parameter parameter distributions.
Results
We successfully trained our model of a single photoreceptor with gradient descent and found phototransduction cascade parameters that fit the electrophysiology data from Chen and colleagues [3], as well as parameters of the ribbon synapse model that fit glutamate release data from Szatko, Korympidou, and colleagues [6]. We then built a network of photoreceptors with these trained parameters and a horizontal cell, and we trained the network’s 200 synaptic conductances to classify images in which contrast and global luminance levels were distorted. The model was able to classify these images despite these distortions, providing further evidence that the structure of the outer retina facilitates contrast normalization.
Discussion
Biophysical models are capable of implementing a variety of computations that are often attributed to larger neural networks higher in the sensory processing hierarchy. For instance, the fitted model of the phototransduction cascade enables a layer of photoreceptors to adapt to drastically different global luminance levels [3] while at the same time regulating glutamate release consistent with data. Our model, fit to multimodal data, can also classify images with different contrasts using very few trainable parameters. This small but biophysically-inspired network may support many other computations as well, broadening our appreciation of the outer retina.

Acknowledgements
Hertie Stiftung, DFG, ERC Starting Grants NextMechMod and DeepCoMechTome)
References
● https://doi.org/10.1016/C2019-0-00467-0
● https://doi.org/10.1101/2024.08.21.608979
● https://doi.org/10.7554/eLife.93795.1
● https://doi.org/10.1016/j.visres.2009.03.003
● https://doi.org/10.7554/eLife.67851
● https://doi.org/10.1038/s41467-020-17113-8

Speakers

Michael Deistler

Kyra Kadhim

Jonas Beck

Ziwei Huang

Philipp Berens

Jakob H Macke

Thomas Euler

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P126: Predictive Coding in the Drosphila Optic Lobe

P129: Of mice and men: Dendritic architecture differentiates human from mice neuronal networks

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P129 Of mice and men: Dendritic architecture differentiates human from mice neuronal networks

Lida Kanari∗1, Ying Shi1,5, Alexis Arnaudon1, Natalı Barros-Zulaica1, Ruth Benavides-Piccione2, Jay S. Coggan1, Javier DeFelipe2, Kathryn Hess3, Huib D. Mansvelder4, Eline J. Mertens4, Julie Meystre5, Rodrigo de Campos Perin5, Maurizio Pezzoli5, Roy T. Daniel6, Ron Stoop7, Idan Segev8, Henry Markram1and Christiaan P.J. de Kock4

1Blue Brain Project, Ecole Polytechnique Federale de Lausanne (EPFL), Geneva, Switzerland.
2Laboratorio Cajal de Circuitos Corticales, Universidad Politecnica de Madrid and Instituto Cajal (CSIC), Madrid, Spain
3Laboratory for Topology and Neuroscience, Brain Mind Institute, Ecole Polytechnique Federale de Lausanne (EPFL), Lausanne, Switzerland
4Department of Integrative Neurophysiology, Center for Neurogenomics and Cognitive Research, Vrije Universiteit Amsterdam, Amsterdam, the Netherlands
5Laboratory of Neural Microcircuitry, ´Ecole Polytechnique Federale de Lausanne (EPFL), Lausanne, Switzerland
6Department of Clinical Neurosciences, Neurosurgery Unit, Centre Hospitalier Universitaire Vaudois, Lausanne, Switzerland
7Center for Psychiatric Neurosciences, Department of Psychiatry, Lausanne University Hospital Center, Lausanne, Switzerland
8Department of Neurobiology and Edmond and Lily Safra Center for Brain Sciences, The Hebrew University of Jerusalem, Jerusalem, Israel

* Email: lida.kanari@gmail.com

Introduction
The organizational principles that distinguish the human brain from other species have been a long-standing enigma in neuroscience. Numerous studies have investigated the correlations between intelligence and neuronal density [1], cortical thickness [2], gyrification [3], and dendritic architecture [4]. However, despite extensive endeavors to unravel its mysteries, numerous aspects of our unique characteristics remain elusive. Along several other factors that contribute in human intelligence, in this study [5] we demonstrate that the shapes of dendrites are an important indicator of network complexity that cannot be disregarded in our quest to identify what makes us human.

Results
Using experimental pyramidal cell reconstructions [6], we built representative mouse and human cortical networks (Fig. 1). We integrate experimental data, taking into account the lower cell density in human cortex layers 2 and 3 [7, 8] and the greater interneuron percentages in the human cortex [9]. Human pyramidal cells form highly complex networks (Fig. 1C), demonstrated by the increased number and simplex dimension compared to mice. Simple dendritic scaling cannot explain species-specific connectivity differences. Topological comparison of dendritic structure reveals much higher perisomatic (basal and oblique) branching density in human pyramidal cells (Fig1. D), impacting network complexity.

Methods
The Topological Morphology Descriptor [10] represents the neuronal morphology as a persistence barcode, using topological data analysis to characterize the shapes of neurons. Scaling transformations were analyzed to compare mouse and human neurons, with optimization via gradient descent. The connectivity was computed using computational modeling of the cortical layers 2 and 3 [11], and approximates the set of potential connections in mouse and human cortex. Memory capacity was analyzed based on dendritic processing models [12].

Discussion
Despite lower neuronal density, human pyramidal cells establish higher-order interactions via their distinct dendritic topology, forming complex networks. This enhanced connectivity is supported by interneurons, which maintain excitation-inhibition balance. The increased dendritic complexity of human pyramidal cells correlates with increased memory capacity, suggesting its role in computational efficiency. Rather than increasing neuron count, human brains prioritize single-neuron complexity to optimize network function. Our findings highlight dendritic morphology as a key determinant of network performance, shaping cognition and future research directions.

Figure 1. Fig1: Multiscale comparison of mouse and human brains, from brain regions to single neurons (A). Greater network complexity (C) emerges in human networks despite the lower neuron density (B), correlating with the higher dendritic complexity of human pyramidal cells. Our findings suggest that dendritic complexity (D) is more substantial for network complexity than neuron density.
Acknowledgements
BBP, EPFL, by ETH Board by SFIT. H.D.M. and C.d.K. by grant awards U01MH114812, UM1MH130981-01 from NIMH, grant no. 945539 (HBP SGA3) Horizon 2020 Framework, NWO 024.004.012, ENW-M2, OCENW.M20.285. R.S. by SNSF (IZLSZ3\_148803, IZLIZ3\_200297, IZLCZ0_206045, 31003A_138526) and Synapsis Foundation (2020-PI02). J.D.F. and R.B.P. by PID2021-127924NB-I00(MCIN/AEI/10.13039/501100011033).

References
[1]https://doi.org/10.3389/neuro.09.031.2009
[2]https://doi.org/10.1016/j.intell.2013.07.010
[3]http://dx.doi.org/10.1016/j.cub.2016.03.021
[4]https://doi.org/10.1016/j.tics.2022.08.012
[5]https://doi.org/10.1101/2023.09.11.557170
[6]https://doi.org/10.1093/cercor/bhv188
[7]https://doi.org/10.1023/A:1024130211265
[8]https://doi.org/10.1023/a:1024134312173
[9]https://doi.org/10.1126/science.abo0924
[10]https://doi.org/10.1007/s12021-017-9341-1
[11]https://doi.org/10.1016/j.cell.2015.09.029
[12]https://doi.org/10.1016/S0896-6273(01)00252-5

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P130: Manifold Inference by Maximising Information: Hypothesis-driven extraction of CA1 neural manifolds via information theory

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P130 Manifold Inference by Maximising Information: Hypothesis-driven extraction of CA1 neural manifolds via information theory

Michael G. Kareithi*1, Mary Ann Go 1, Pier Luigi Dragotti 2 Simon R. Schultz1

1 Department of Bioengineering, Imperial College London, London, United Kingdom
2Department of Electrical and Electronic Engineering, Imperial College London, London, United Kingdom

*Email: m.kareithi21@imperial.ac.uk

Introduction
Neural manifolds have been a useful concept for understanding cognition, with recent work showing the importance of "hypothesis-driven" analyses: linking behaviour with manifolds via supervised manifold learning [1]. Linear dimensionality reduction methods are easier to interpret than their nonlinear counterparts, but often can only detect linear correlations in neural activity. From an information-theory perspective, a natural approach to supervised manifolds is to maximise Mutual Information between the embedding and the target variable. We use simple linear embeddings with an information-theoretic objective function: Quadratic Mutual Information [2], and apply it as a tool for hypothesis-driven manifold learning in mouse hippocampus.
Methods
Quadratic Mutual Information (QMI) is derived from Renyi entropy and divergence, a broader family of measures than Shannon entropy and mutual information (MI), the latter being a special case of the former. Like MI, QMI has the desirable property of being zero if and only if the variables are independent. Its advantage is that it can be estimated with high-dimensional data, and is differentiable. We fit a linear projection from activity to a lower-dimensional subspace by maximising QMI between the projection and a target variable. We call our framework Manifold Inference by Maximising Information (MIMI). We apply MIMI to two-photon calcium recordings in mouse CA1 during a 1D running task.
Results
In our dataset, mice run continuously along a circular track [3]. We fit MIMI on calcium fluorescence activity, with the animal's angular position as target variable, cross-validating with a 75%-25% train-test split. In four out of eight mice we find the majority of position-information in a 2-3 dimensional subspace (Fig 1.a). In the sessions without informative subspaces, the linear decodability of position is low even from the full population activity, indicating the absence of a population code (Fig 1.e). The informative subspaces contain ring-shaped manifolds mapping continuously onto the animal's physical coordinates (Fig 1.f).
Discussion
Combining information-theoretic measures with linear embeddings is a useful idea for analysing populations, where our aim is not only to find manifold structure, but to understand how cell assemblies coordinate to sculpt it. MIMI shows that we can find behaviourally-informative manifolds without nonlinear embeddings: only a nonlinear measure of dependence. Downstream analysis can then pose questions about representation by examining the linear transformation weights: for example, asking if two variables are represented orthogonally. We believe MIMI will be a useful framework for interpretable, hypothesis-driven manifold analysis.

Figure 1. A) Explained position variance (R-squared of ridge-regressor, left) and Mutual Information (right) between position and MIMI projection at different dimensionalities. Each line is an individual mouse. B) Position-variance explained by full population vs by MIMI subspace. C) Activity in MIMI subspace for four mice with informative subspaces, coloured by associated position of mouse.
Acknowledgements-
References
1.https://doi.org/10.1038/s41586-023-06031-6
2.https://doi.org/10.1007/978-1-4419-1570-2_2
3.https://doi.org/10.3389/fncel.2021.618658

Speakers

Mary Ann Go

Simon R Schultz

Michael Kareithi

Pier Luigi Dragotti

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P131: Personalized Computational Models for Selective and Impact-Driven Brain Stimulation

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P131 Personalized Computational Models for Selective and Impact-Driven Brain Stimulation

Fariba Karimi*1,2,Taylor Newton1,Melanie Steiner1, Antonino Cassara1,Niels Kuster1,2 ,Esra Neufeld1

1IT’IS Foundation, Zürich, Switzerland
2Swiss Federal Institute of Technology (ETH Zurich), Zürich, Switzerland
3Ecole Polytechnique Fédérale de Lausanne (EPFL), Geneva and Sion, Switzerland
4Clinical Neuroscience, University Medical School of Geneva, Geneva, Switzerland

*Email: karimi@itis.swiss

Introduction

Non-invasive brain stimulation (NIBS) offers promising therapeutic avenues for a range of neurological conditions. However, inter-subject response variability remains an important challenge, often limiting its widespread clinical adoption. Here, we present a computational pipeline designed to optimize NIBS by harnessing personalized brain network dynamics modeling, towards enhancing both effictivity and predictability of therapeutic outcomes.

Methods
We developed a comprehensive pipeline on the o2S2PARC platform (see Fig. 1). The pipeline utilizes MRI and diffusion-weighted imaging (DWI) data to construct detailed head models (>40 distinct tissue types) through AI segmentation, perform electromagnetic (EM) simulations to determine exposure-induced electric fields and personalized lead field matrices, and predict the impact of diverse stimulation conditions on brain network dynamics using personalized neural mass models (NMMs; derived from DWI structural connectivity data; simulated using the The Virtual Brain (TVB) [1] framework). The brain network modeling combined with the personalized lead fields permit to synthetize virtual EEG signals that can be compared with measurable data.
Results
Using the developed pipeline, we implemented a temporal interference stimulation planning (TIP) tool for optimizing electrode locations for temporal interference stimulation (TIS, a recently introduced transcranial electric stimulation method capable of targeted stimulation at depth). Demonstration applications of our pipeline predicted shifts in EEG spectral responses following transcranial alternating current stimulation (tACS) in accordance with theoretical and empirical data. Additionally, our simulations revealed dynamic fluctuations of inter-hemispheric synchronization in accordance with experimental observations. These results underscore our pipeline's potential in modeling real-world brain responses to NIBS [3].

Discussion
We established a fully automated computational pipeline for personalized NIBS modeling and the optimization of dynamic brain network response predictions. This pipeline underscores the shift from generic exposure-targeting approaches to a personalized, impact-driven (network dynamics) approach, towards improving the efficacy and precision of NIBS therapies. Current research focuses on the continuous inference of improved model parameters based on measurement feedback and model-predictive control. This works lays the groundwork for adaptive and effective brain dynamics modulation for the treatment of complex neurological disorders, marking a significant advance in the personalized medicine landscape [3].

Figure 1. Figure 1: Schematic representation of the developed pipeline on the o2S2PARC platform
Acknowledgements
--
References
1.https://doi.org/10.1016/j.neuroimage.2015.01.002
2.https://doi.org/10.1109/TNSRE.2012.2200046
3.https://doi.org/10.1088/1741-2552/adb88f

Speakers

Antonino Cassara

Esra Neufeld

Fariba Karimi

Taylor Newton

Melanie Steiner

Niels Kuster

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P132: Brain Symphony: A Transformer-Driven Fusion of fMRI Time Series and Structural Connectivity

P135: Neural dynamics of reversal learning in the prefrontal cortex and recurrent neural networks

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P135 Neural dynamics of reversal learning in the prefrontal cortex and recurrent neural networks

Christopher M. Kim*1, Carson C. Chow1, Bruno B. Averbeck2

1Laboratory of Biological Modeling, NIDDK/NIH, Bethesda, MD
2Laboratory of Neuropsychology, NIMH/NIH, Bethesda, MD
3Current address: Department of Mathematics, Howard University, Washington, DC

*Email: christopher.kim@howard.edu

Introduction

In a probabilistic reversal learning task, a subject learns from initial trials that one of the two options yields reward with higher probability than the other (for instance, the high-value and the low-value options are rewarded 70% and 30% of the time, respectively). When the reward probabilities of two options are reversed at a random trial, the agent must switch its choice preference to maximize reward. Such reversal learning has been used for assessing one’s ability to adapt in a dynamically changing environment with uncertain rewards [1]. In this task, reward outcomes must be integrated over multiple trials before reversing the preferred choice, as the less favorable option yields rewards stochastically.

Methods
We investigated how cortical neurons represent integration of decision-related evidence across trials in the reversal learning task. Previous works considered attractor dynamics along a line in the state space as a neural mechanism for evidence integration [2]. However, when integrating evidence across trials, the subject must perform task-related behaviors within each trial, which could induce non-stationary neural activity. To understand the neural representation of multi-trial evidence accumulation, we analyzed the activity of neurons in the prefrontal cortex of monkeys and recurrent neural networks trained to perform a reversal learning task.
Results
We found that, in a neural subspace encoding reversal probability, its activity represented integration of reward outcomes as in a line attractor. The reversal probability activity at the start of a trial was stationary, stable and consistent with the attractor dynamics. However, during the trial, the activity was associated with task-related behavior and became non-stationary, thus deviating from the line attractor. Fitting a predictive model to neural data showed that the stationary state at the trial start served as an initial condition for launching the non-stationary activity. This suggested an extension of the line attractor model with behavior-induced non-stationary dynamics.
Discussion
Our findings show that, when performing a reversal learning task, a cortical circuit represents reversal probability, not only in stable stationary states as in a line attractor model, but also in dynamic neural trajectories that can accommodate non-stationary task-related behaviors necessary for the task. Such neural mechanism demonstrates the temporal flexibility of cortical computation and opens the opportunity for extending existing neural model for evidence accumulation by augmenting temporal dynamics.

Acknowledgements
This research was supported by the Intramural Research Program of the National Institutes of Health: the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) and the National Institute of Mental Health (NIMH). This work utilized the computational resources of the NIH HPC Biowulf cluster (https://hpc.nih.gov).
References
[1] Bartolo, R., & Averbeck, B. B. (2020). Prefrontal cortex predicts state switches during reversal learning.Neuron,106(6), 1044-1054.
[2] Mante, V., Sussillo, D., Shenoy, K. V., & Newsome, W. T. (2013). Context-dependent computation by recurrent dynamics in prefrontal cortex.nature,503(7474), 78-84.

Speakers

Christopher Kim

Carson Chow

Bruno Averbeck

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P136: Computational Modeling of Ca2+ Blocker Effect in the Thalamocortical Network of Epilepsy: A Dynamic Causal Modeling Study

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P136 Computational Modeling of Ca2+ Blocker Effect in the Thalamocortical Network of Epilepsy: A Dynamic Causal Modeling Study

Euisun Kim1, Jiyoung Kang2, Jinseok Eo3, Hae-Jeong Park*1,3,4,5

¹Graduate School of Medical Science, Brain Korea 21 Project, Yonsei University College of Medicine, Seoul, Republic of Korea
²Department of Scientific Computing, Pukyong National University, Busan, Republic of Korea
³Center for Systems and Translational Brain Sciences, Institute of Human Complexity and Systems Science, Yonsei University, Seoul, Republic of Korea
4Department of Cognitive Science, Yonsei University, Seoul, Republic of Korea
5Department of Nuclear Medicine, Severance Hospital, Yonsei University College of Medicine, Seoul, Republic of Korea
*1,2 are equally contributed.

*Email: parkhj@yonsei.ac.kr
Introduction

Childhood Absence Epilepsy (CAE) is characterized by excessive thalamocortical synchronization, leading to recurrent loss of consciousness [1]. This phenomenon is linked to T-type calcium channel hyperactivity, a key driver of seizure generation [2]. Ethosuximide (ETX), a T-type Ca²⁺ blocker and first-line CAE treatment, is expected to influence both intrinsic neural property and interregional connectivity, but its mechanism on thalamocortical network hierarchy remain unclear. This study employs Dynamic Causal Modeling (DCM) to analyze ETX-induced network changes from a neuropharmacological perspective [3].

Methods
To examine ETX-induced changes in thalamocortical dynamics, we incorporated voltage dependent calcium channels to a thalamocortical model (TCM) [4]. The model included six cortical populations (pyramidal, interneuron, and stellate cells) and two thalamic populations (reticular and relay neurons) for a thalamocortical system. Their temporal evolution is governed by coupled differential equations, describing membrane potential and conductance changes mediated by AMPA, NMDA, GABA-A receptors, and T-type calcium channels —the latter capturing ETX effects. Resting-state EEG data were collected before and after ETX administration in CAE patients. Using DCM of longitudinal EEG, we analyzed hierarchical thalamocortical connectivity changes and modeled nonlinear interactions influencing EEG cross-spectral density (CSD) within the Default Mode Network (DMN), including the mPFC, Precuneus, and lateral parietal cortices —which is often aberrantly deactivated during CAE seizures, potentially due to subcortical inhibition [5].
Results
ETX significantly altered both thalamocortical and cortical network dynamics. We observed changes in intrinsic neural properties as well as interregional connectivity when comparing pre- and post-ETX conditions. These findings indicate that ETX modulates local neural excitability and large-scale network interactions, thereby contributing to seizure suppression in CAE.

Discussion
By incorporating voltage-dependent Ca²⁺ channels into a thalamocortical model, this study provides a preliminary computational evidence that calcium channel blockers help restore large-scale network stability in CAE. The results underscore the therapeutic mechanism by which these agents modify pathological thalamocortical interactions. Further validation and refinement of the computational model may enhance clinical approaches to treating CAE and related epileptic disorders.

Acknowledgements
This research was supported by the Bio&Medical Technology Development Program of the National Research Foundation (NRF) funded by the Korean government (MSIT) (No. RS-2024-00401794).
References
● https://doi.org/10.1016/j.nbd.2023.106094
● https://doi.org/10.1111/epi.13962
● https://doi.org/10.1016/j.neuroimage.2023.120161
● https://doi.org/10.1016/j.neuroimage.2020.117189
● 10.3233/BEN-2011-0310

Speakers

Euisun Kim

Jiyoung Kang

Jinseok Eo

Hae-Jeong Park

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P137: Coordinated Multi-frequency Oscillatory Bursts Enable Time-structured Dynamic Information Transfer

P140: Single-unit responses to dynamic salient visual stimuli in the human medial temporal lobe

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P140 Single-unit responses to dynamic salient visual stimuli in the human medial temporal lobe

Alina Kiseleva1, Eva van Gelder1,Hennric Jokeit1, Johannes Sarnthein2, Lukas Imbach1, Tena Dubcek1& Debora Ledergerber*11Swiss Epilepsy Clinic, Clinical Neurophysiology, Zürich, Switzerland (Debora.Ledergerber@kliniklengg.ch)2Universitätsspital Zürich, Klinik für Neurochirurgie, Zürich, Switzerland

*Email:Debora.Ledergerber@kliniklengg.ch
Introduction

The medial temporal lobe (MTL) is critical for mnemonic functions, navigation and social cognition. For many of these higher-order cognitive processes, correlates of single-neuron responses have been found in different regions of human MTL. Amygdala neurons respond to emotional stimuli [1], while hippocampus (HC) and entorhinal cortex (EC) neurons encode memory [2] and navigation [3]. Efficient encoding of task covariates depends on neurons with mixed selectivity, found in rodent subiculum and EC [4]. While this coding scheme has been described in human MTL [5], it remains elusive whether it is applied differentially in different contexts.

Methods
We investigated the activity of 500 neurons in human MTL while participants watched a movie with alternating neutral and emotionally charged clips [6]. To model neuronal firing rates, we applied a Generalized Linear Model, using three covariates: trial type (Face/Landscape), size of the dominant object, and its movement across frames. We then implemented a model selection procedure to identify neurons specifically tuned to each covariate.
Results
We found the highest number of neurons encoding the difference between trials of landscapes versus emotional faces (14%). A smaller but substantial population of neurons showed specificity for the main object size and degree of movement (5% and 6%). Additionally, 3% of neurons demonstrated mixed selectivity, responding to the combination of at least two visual features.
Despite the amygdala's established role in processing of emotional stimuli, we found only a slightly increased number of neurons specific to emotional trials in the amygdala compared to HC and EC, and the difference in the proportion of emotionally responsive neurons across the MTL was not statistically significant (P > 0.9, χ² test).
Discussion
Overall, it suggests emotional stimulus processing is distributed across MTL regions and neurons encoding emotional stimuli may additionally show selectivity for other task features. The presence of mixed selectivity further highlights the integrative role of MTL neurons in processing complex visual and emotional information, potentially supporting flexible cognitive functions.

Acknowledgements
We sincerely appreciate the time and contribution of all patients who participated in this study. We are also grateful to our colleagues and collaborators for their insightful discussions and support. We extend our deep gratitude to the clinical staff for their invaluable assistance in data collection.
References
1.https://doi.org/10.1073/pnas.1323342111
2.https://doi.org/10.1523/jneurosci.1648-20.2020
3.https://doi.org/10.1038/nn.3466
4.https://doi.org/10.1016/j.celrep.2021.109175
5.https://doi.org/10.1016/j.celrep.2024.114071
6.https://doi.org/10.1038/s41597-020-00790-x

Speakers

Alina Kiseleva

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P142: Finite-sampling bias correction for discrete Partial Information Decomposition

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P142 Finite-sampling bias correction for discrete Partial Information Decomposition

Loren Koçillari*1,2, Gabriel M. Lorenz1,4, Nicola M. Engel1, Marco Celotto1,5, Sebastiano Curreli3, Simone B. Malerba1, Andreas K. Engel2, Tommaso Fellin3, and Stefano Panzeri1
1Institute for Neural Information Processing, Center for Molecular Neurobiology, University Medical Center Hamburg-Eppendorf (UKE), Hamburg, Germany
2Department of Neurophysiology and Pathophysiology, University Medical Center Hamburg-Eppendorf (UKE), Hamburg, Germany
3Istituto Italiano di Tecnologia, Genova, Italy
4Department of Pharmacy and Biotechnology, University of Bologna, Bologna, Italy
5Department of Brain and Cognitive Sciences, Picower Institute for Learning and Memory, Massachusetts Institute of Technology, Cambridge, Massachusetts, USA
*Email: l.kocillari@uke.de
Introduction

A major question in neuroscience is how groups of neurons interact to generate behavior. Shannon Information Theory has been widely used to quantify dependencies among neural units and cognitive variables [1]. Partial Information Decomposition (PID) [2,3] has extended Shannon theory to decompose neural information into synergy, redundancy, and unique information. Discrete versions of PID are suitable for spike train analysis. However, estimating information measures from real data is subject to a systematic upward bias due to limited sampling [4], an issue that has been largely overlooked in PID analyses of neural data.

Methods
Here, we first studied the bias of discrete PID through simulations of neuron pairs with varying degrees of synergy and redundancy, using sums of Poisson processes with individual and shared terms modulated by stimuli. We assumed that the bias of union information (the sum of unique information and redundancy) equals that of the information obtained from stimulus-uncorrelated neurons. We found that this assumption accurately matched simulated data, allowing us to derive analytical approximations of PID biases in large sample sizes. We used this knowledge to develop efficient bias-correction methods, validating them on empirical recordings from 53,113 neuron pairs in the auditory cortex, posterior parietal cortex, and hippocampus of mice.
Results
Our results show that limited sampling bias affects all terms in discrete PIDs, with synergy exhibiting the largest upward bias. The bias of synergy grows quadratically with the number of possible discrete responses of individual neurons, whereas the bias of unique information scales linearly and has intermediate values, while redundancy remains almost unbiased. Thus, neglecting or failing to correct for this bias leads to substantially inflated synergy estimates. Simulations and real data analyses showed that our bias-correction procedures can mitigate this problem, leading to much more precise estimates of all PID components.

Discussion
Our study highlights the systematic overestimation of synergy in both simulated and empirical datasets, underscoring the need for bias-correction methods, and offers empirically validated ways to correct for this problem. These findings provide a computational and theoretical basis for enhancing the reliability of PID analyses in neuroscience and related fields. Our work informs experimental design by providing guidelines on the sample sizes required for unbiased PID estimates and supports computational neuroscientists in selecting efficient PID bias-correction methods.

Acknowledgements
This work was supported by the NIH Brain Initiative grant U19 NS107464 (to SP and TF), the NIH Brain Initiative grant R01 NS109961 and R01 NS108410, the Simons Foundation for Autism Research Initiative (SFARI) grant 982347 (to SP), the European Union’s European Research Council grants NEUROPATTERNS 647725 (to TF) and cICMs ERC-2022-AdG-101097402 (to AKE).
References
1.Quian Quiroga, R, Panzeri, S (2009). Extracting information from neuronal populations: information theory and decoding approaches.Nature Reviews Neuroscience, 10, 173-185.
2.Williams, PL, Beer, RD (2010). Nonnegative decomposition of multivariate information.arXiv preprintarXiv:1004.2515.
3.Bertschinger, N, Rauh, J, Olbrich, E, Jost, J, Ay, N. (2014). Quantifying unique information.Entropy, 16, 2161-2183.
4.Panzeri, S, Treves, A (1996). Analytical estimates of limited sampling biases in different information measures.Network: Computation in neural systems, 7, 87-107.

Speakers

Loren Kocillari

Andreas K. Engel

Tommaso Fellin

Gabriel M. Lorenz

Nicola M. Engel

Marco Celotto

Sebastiano Curreli

Simone B. Malerba

Stefano Panzeri

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P143: Redundant stimulus encoding in ferret cortex during a lateralized detection task

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P143 Redundant stimulus encoding in ferret cortex during a lateralized detection task

Loren Koçillari*1,2, Edgar Galindo-Leon2, Florian Pieper2, Stefano Panzeri1, Andreas K. Engel2
1Institute for Neural Information Processing, Center for Molecular Neurobiology, University Medical Center Hamburg-Eppendorf (UKE), Hamburg, Germany

2Department of Neurophysiology and Pathophysiology, University Medical Center Hamburg-Eppendorf (UKE), 20246 Hamburg, Germany

*Email: l.kocillari@uke.de
Introduction

The brain’s ability to integrate diverse sources of information is crucial for perception and decision-making. It can combine inputs synergistically to increase information capacity or redundantly to enhance signal reliability and robustness. Previous research has shown that redundant information between mouse auditory neurons increases during correct compared to incorrect trials in a tone discrimination task [1]. However, it remains unclear how redundancy’s behavioral role generalizes at larger scales, across frequency bands, and between unimodal and multimodal sensory stimuli. Using Partial Information Decomposition (PID) [2], we analyze redundant and synergistic information in ferret cortical activity during an audiovisual task.

Methods
We studied information processing in behaving ferrets during a visual or audiovisual stimulus detection task [3]. Brain activity from auditory, visual, and parietal areas of the left hemisphere was recorded using a 64-channel ECoG array [3]. We quantified task-related changes in single-channel local field potential (LFP) power and phase across time and frequency bands. We assessed stimulus encoding in individual channels by computing time-resolved Shannon mutual information between stimulus location and LFP power or phase. Finally, using PID, we quantified behaviorally relevant synergistic and redundant stimulus-related information conveyed by channel pairs at information peaks, in relation to correct choices and faster reaction times.

Results
We found that stimulus information, for both LFP power and phase, was primarily present in the peri-stimulus interval at lower frequency bands (theta and alpha), while beta and gamma bands contained less information. Stimulus information in the theta band was greater in hit trials than in miss trials and in fast-hit trials than in slow-hit trials, suggesting that the information content of theta activity is behaviorally relevant. Redundancy across channel pairs in the theta-band was higher in hit than in miss trials and in fast-hit trials than in slow-hit trials, whereas synergy was greater in miss and slow-hit trials.

Discussion
Our results suggest that the amount of information encoded in the theta band is behaviorally relevant for perceptual discrimination. They also indicate that redundancy is more beneficial than synergy for correct or rapid perceptual judgements during both visual and audiovisual stimulus detection. This supports the notion that the advantages of redundancy in downstream signal propagation and robustness outweigh its limitations of the total information that can be encoded across areas.

Acknowledgements
This work was supported by the cICMs ERC-2022-AdG-101097402 (to AKE).
References
1.Koçillari, L., et al. (2023). Behavioural relevance of redundant and synergistic stimulus information between functionally connected neurons in mouse auditory cortex.Brain Informatics,10(1), 34.
2.Williams, PL, Beer, RD (2010). Nonnegative decomposition of multivariate information.arXiv preprintarXiv:1004.2515.
3.Galindo-Leon, E. E., et al. (2025). Dynamic changes in large-scale functional connectivity prior to stimulation determine performance in a multisensory task.Frontiers in Systems Neuroscience,19, 1524547.

Speakers

Loren Kocillari

Andreas K. Engel

Stefano Panzeri

Edgar Galindo-Leon

Florian Pieper

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P144: Event-driven eligibility propagation: combining efficiency with biological realism

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P144 Event-driven eligibility propagation: combining efficiency with biological realism

Agnes Korcsak-Gorzo*1,2, Jesús A. Espinoza Valverde3, Jonas Stapmanns4, Hans Ekkehard Plesser5,1,6, David Dahmen1, Matthias Bolten3, Sacha J. van Albada1,7, Markus Diesmann1,2,8,9
1Institute for Advanced Simulation (IAS-6), Computational and Systems Neuroscience, Forschungszentrum Jülich, Jülich, Germany
2Fakultät 1, RWTH Aachen University, Aachen, Germany
3Department of Mathematics and Science, University of Wuppertal, Wuppertal, Germany
4Department of Physiology, University of Bern, Bern, Switzerland
5Department of Data Science, Faculty of Science and Technology, Norwegian University of Life Sciences, Aas, Norway
6Käte Hamburger Kolleg, RWTH Aachen University, Aachen, Germany
7Institute of Zoology, University of Cologne, Cologne, Germany
8JARA-Institute Brain Structure-Function Relationships (INM-10), Forschungszentrum Jülich, Jülich, Germany
9Department of Psychiatry, Psychotherapy and Psychosomatics, Medical Faculty, RWTH Aachen University, Aachen, Germany

*Email: a.korcsak-gorzo@fz-juelich.de
Introduction

Understanding the neurobiological computations underlying learning is enhanced by simulations, which serve as a critical bridge between experimental findings and theoretical models. Recently, several biologically plausible learning algorithms have been proposed for simulating spiking recurrent neural networks, achieving performance comparable to backpropagation through time (BPTT) [1]. In this work, we adapt one such learning rule, eligibility propagation (e-prop) [2], to the spiking neural network simulator (NEST) optimized for large-scale simulations.

Methods
To improve computational efficiency and enable large-scale simulations, we replace the original time-driven synaptic updates - executed at every time step - with an event-driven approach, where synapses are updated only when activated by a spike. This requires storing the e-prop history between weight updates, and with optimized history management, we significantly reduce computational overhead. Additionally, we replace components inspired by machine learning with biologically plausible mechanisms and extend the model with features such as continuous dynamics, strict locality, sparse connectivity, and approximations that eliminate vanishing terms, further enhancing computational efficiency.
Results
We demonstrate that our event-driven weight update scheme accurately reproduces the behavior of the original time-driven e-prop model (see Fig. 1) while significantly reducing computational costs, particularly in biologically realistic settings with sparse activity. We validate this approach on various biologically motivated regression and classification tasks, including neuromorphic MNIST [3]. Furthermore, we show that learning performance and computational efficiency remain comparable to those of the original model, despite the incorporation of biologically inspired features. Strong and weak scaling experiments confirm the robust scalability of our implementation, supporting networks with up to millions of neurons.
Discussion
By integrating biologically enhanced e-prop plasticity into an established open-source spiking neural network simulator with a broad and active user base, we aim to facilitate large-scale learning experiments. Additionally, this work provides a foundation for implementing other three-factor learning rules from the extensive literature in an event-driven manner. By bridging AI and computational neuroscience, our approach has the potential to enable large-scale AI networks to leverage energy-efficient biological mechanisms.

Figure 1. Implementation of event-driven e-prop demonstrated on a temporal pattern generation task. Learning occurs through updates to input, recurrent, and output synapses. The upper middle plot illustrates the correspondence between the event-driven and time-driven e-prop models.
Acknowledgements
This work was supported by Joint Lab SMBH; HiRSE_PS; NeuroSys (Clusters4Future, BMBF, 03ZU1106CB); EU Horizon 2020 Framework Programme for Research and Innovation (945539, Human Brain Project SGA3) and Europe Programme (101147319, EBRAINS 2.0); computing time on JURECA (JINB33) via JARA Vergabegremium at FZJ; and Käte Hamburger Kolleg: Cultures of Research (c:o/re), RWTH Aachen (BMBF, 01UK2104).
References
1. Werbos, P. (1990). Backpropagation through time: what it does and how to do it. Proceedings of the IEEE, 78(10), 1550–1560.
2. Bellec, G., Scherr, F., Subramoney, A., Hajek, E., Salaj, D., Legenstein, R., & Maass, W. (2020). A solution to the learning dilemma for recurrent networks of spiking neurons. Nature Communications, 11(1), 3625.
3. Orchard, G., Jayawant, A., Cohen, G., & Thakor, N. (2015). Converting Static Image Datasets to Spiking Neuromorphic Datasets Using Saccades. Frontiers in Neuroscience, 9, 437.

Speakers

Agnes Korcsak-Gorzo

Hans Ekkehard Plesser

David Dahmen

Sacha van Albada

Markus Diesmann

Jesús A. Espinoza Valverde

Jonas Stapmanns

Matthias Bolten

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P145: Biophysical thalamic neuron models to probe the impact of ultrasound induced heating in the brain

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P145 Biophysical thalamic neuron models to probe the impact of ultrasound induced heating in the brain

Rikinder Kour1, Ayesha Jameel2,3, Joely Smith3,4, Peter Bain5,6, Dipankar Nandi5,6, Brynmor Jones3, Rebecca Quest3,4, Wladyslaw Gedroyc2,3, Roman Borisyuk7,Nada Yousif1*

1School of Physics Engineering and Computer Science, University of Hertfordshire, UK
2Department of Surgery and Cancer, Imperial College London, UK
3Department of Imaging, Imperial College Healthcare NHS Trust, London, UK
4Department of Bioengineering, Imperial College London, UK
5Division of Brain Sciences, Imperial College London, UK
6Department of Neurosciences, Imperial College Healthcare NHS Trust, London, UK
7Department of Mathematics and Statistics, University of Exeter, Exeter, UK

* Email: n.yousif@herts.ac.uk
Introduction

High intensity focussed ultrasound (HIFU) is used for ablating thalamic neurons to treat tremor [1]. Low intensity focussed ultrasound (LIFU) can be used for neuromodulation [2] and previous modelling suggests that LIFU induces neuronal excitation via mechanical modulation of the cell membrane [3,4]. Although modelling of neural effects of HIFU is limited, understanding the effects of heating during HIFU at sub-ablative temperatures is important, as this is used for monitoring side effects and clinical improvement during tremor treatment [5]. Here we modified biophysical thalamocortical neuron models [6,7] to look at the change in firing patterns as HIFU induced heating approaches ablative temperatures.

Methods
First, we used data from magnetic resonance thermography performed during a HIFU treatment to select the temperature value for the ‘celsius’ parameter in NEURON [8]. We then examined the effect of temperature on the neuronal firing, as mediated by the parameters of gating equations [9]. Next, we added temperature dependence for the membrane capacitance, as shown experimentally [10] and in a previous modelling study [11]. We compared the effect of temperature in single neurons with one, three and 200 compartments under current clamp conditions with different input current levels [6]. Finally, we considered the impact of increasing temperature on a small network of two excitatory thalamic neurons [7] and two inhibitory reticular neurons.
Results
The thermography data (Fig. 1A) shows that at the HIFU target site, the temperature increased up to 62°C for a treatment sonication. With temperature dependent parameters of the gating equations, increasing temperatures lead to inhibition of the neuron (Fig. 1B). Interestingly, when including a temperature dependent membrane capacitance, we observed a similar pattern of results. Furthermore, we also saw the same effect of temperature on firing rate regardless of the number of compartments modelled. Finally, the network model showed that although with changing temperature the firing of the individual neurons both increased and decreased, we still observe an overall termination of firing in all neurons as the temperature exceeds 40°C.

Discussion
HIFU is commonly used to thermally ablate the thalamus and suppress tremor, via application of ultrasound energy called sonications. Test sonications are used to heat the tissue to sub-ablative temperatures to confirm targeting and test for adverse effects. This study looked at the impact of such sub-ablative heating on single neuron models and a small network representative of the target region. Our results indicate that once temperatures exceed 40°C neuronal firing is completely inhibited. Future work will extend the network model to look at downstream effects of heating. Such work will allow us to better understand the link between subablative temperature increases, suppression of tremor and adverse effects for optimising treatment.

Figure 1. Figure1: (A) The heating induced in by a HIFU treatment sonication. The target is at the centre of the image and the temperature reaches 62°C. (B) The results from simulating a single compartment thalamocortical neuron at different temperatures, when the neuron has only temperature dependent gating equations (black) and when the membrane capacitance has temperature dependence (red).
Acknowledgements
NY is funded by the Royal Academy of Engineering and the Leverhulme Trust and AJ is partially funded by Funding Neuro.
References[1] 10.3389/fneur.2021.654711 [2] 10.1016/j.cub.2013.10.029 [3] 10.1523/ENEURO.0136-15.2016 [4] 10.1088/1741-2552/ab1685 [5] doi.org/10.1002/ana.26945 [6] 10.1523/JNEUROSCI.18-10-03574.1998 [7] 10.1152/jn.1996.76.3.2049 [8] 10.1007/978-1-4614-7320-6_795-2 [9] 10.1007/978-1-4614-7320-6_236-1 [10] 10.1016/0301-4622(94)00103-Q [11] 10.3389/fncom.2022.933818

Speakers

Nada Yousif

Rikinder Kour

Ayesha Jameel

Joely Smith

Peter Bain

Dipankar Nandi

Brynmor Jones

Rebecca Quest

Wladyslaw Gedroyc

Roman Borisyuk

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P146: Fast Visual Reorientation in Postsubicular Head-Direction Cells Conditional on Cue Visibility

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P146 Fast Visual Reorientation in Postsubicular Head-Direction Cells Conditional on Cue Visibility

Sven Krausse1,2, Emre Neftci1,2,Alpha Renner*1
1Forschungszentrum Jülich, Aachen, Germany
2RWTH Aachen, Aachen, Germany

*Email: a.renner@fz-juelich.de
Introduction

Accurate spatial navigation relies on head-direction (HD) cells, which encode orientation in allocentric coordinates, like a neural compass [1,2]. Found, e.g., in postsubiculum (PoSub) and thalamus, HD cells integrate angular velocity signals from vestibular, proprioceptive, and optic flow inputs, recalibrating via visual cues [2] to avoid drift. Reorientation speed after cue absence is key to understanding the HD system’s dynamics and for bio-inspired models.[3]reported rapid reorientation, while[4]suggested an internal gain factor modulates it, though its mechanism remains unclear. Using a new dataset [5], we examine reorientation dynamics, finding it is fast but contingent on cue visibility.
Methods
We analyzed a dataset [5] containing head tracking and PoSub spike trains from six mice. Internal HD was decoded from spikes using a Bayesian approach [5]. Mice navigated a circular platform with dim LED cues (Fig. 1a) alternating between adjacent walls in 16 trials.Trials were excluded if >20% of the first minute after a cue switch had unreliable tracking, if movement ceased for >5 s, or if HD failed to reorient. Using head tracking data, we reconstructed each mouse’s visual field (FOV = 180°) to estimate cue visibility. Reorientation speed was quantified via exponential fits (scipy.optimize.curve_fit). Time constants (τ) were constrained to 0.1–3 s, with magnitude limits of 0–90°. Aligned mean error fits (Figs. 1b,c) estimated unconstrained τ, magnitude, and no delay.
Results
In Fig. 1d, after a cue switch, decoding error decreases from 90° as HD reorients. Reorientation does not always occur immediately but around when the cue becomes visible. Comparing error aligned in time by cue switch (Fig. 1b) vs. fitted delay (1c), the latter improves alignment and yields faster τ. Fig. 1e suggests that fitted switching times can be predicted from the mouse’s FOV, but only for “reorientation” trials (blue) where the cue appeared outside the FOV. Cues appearing within the FOV may cause a conflict between reanchoring and reorientation due to the lack of a dark phase between trials. Prediction cannot be perfect as pupil orientation and blinking are unknown. Based on these preliminary results we develop a model of reorientation dynamics to capture additional effects.
Discussion
Consistent with [3], we confirm that reorientation occurs in abrupt jumps, but alignment must consider visual FOV rather than assuming omnidirectional vision. While in [3]. mice were trained to fixate cues, FOV’s role may seem trivial but is often ignored. Our findings offer a better mechanistic understanding of the gain factor that mediates reorientation speed found by [4] in thalamus, which is not yet mechanistically explained. More broadly, our results contribute to an integrative model of HD reorientation and reanchoring, advancing both neuroscientific understanding and bio-inspired navigation systems (which we plan to build in the future [6]).

Figure 1. Fig. 1 a. Arena, platform, cues and FOV b. Decoding error aligned by cue switch c. Error aligned by fitted internal HD switch d. Single trial where cue switch occurs roughly as the cue enters FOV. Difference between red and black curves is decoding error (blue). e. Estimated time until cue becomes visible vs. fitted delay. Diagonal in black, points where cue appears within FOV in grey.
Acknowledgements
This research was funded by VolkswagenStiftung [CLAM 9C854]. For this work, the data from Duszkiewicz et al. (2024) [1] was used, and we thank the authors for making this data available. We especially thank Adrian Duszkiewicz for answering our questions and providing additional advice on the data. We thank Johannes Leugering, Friedrich Sommer and Paxon Frady for their feedback.
References
[1] Rank, J. B. (1984). Head-direction cells in the deep layers of dorsal presubiculum of freely moving rats. In Soc. Neuroscience Abstr. (Vol. 10, p. 599).
[2] Taube et al. (1990).https://doi.org/10.1523/JNEUROSCI.10-02-00420.1990
[3] Zugaro et al. (2003).https://doi.org/10.1523/JNEUROSCI.23-08-03478.2003
[4] Ajabi et al. (2023).https://doi.org/10.1038/s41586-023-05813-2
[5] Duszkiewicz et al. (2024).https://doi.org/10.1038/s41593-024-01588-5
[6] Krausse et al. (2025).https://doi.org/10.48550/arXiv.2503.08608

Speakers

Sven Krausse

Emre Neftci

Alpha Renner

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P147: Latency correction in sparse neuronal spike trains with overlapping global events

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P147 Latency correction in sparse neuronal spike trains with overlapping global events

Arturo Mariani1, Federico Senocrate1, Jason Mikiel-Hunter2, David McAlpine2, Barbara Beiderbeck3,

Michael Pecka4, Kevin Lin5,Thomas Kreuz6,7∗

1Department of Physics and Astronomy, University of Florence, Sesto Fiorentino, Italy

2Department of Linguistics, Macquarie University, Sydney, Australia

3 Graduate School of Systemic Neurosciences, Ludwig-Maximilians-Universität, Munich, Germany

4 Division of Neurobiology, Faculty of Biology, Ludwig-Maximilians-Universität, Munich, Germany

5École Nationale Supérieure de l’Électronique et de ses Applications, Cergy, France

6Institute for Complex Systems (ISC), National Research Council (CNR), Sesto Fiorentino, Italy

7National Institute of Nuclear Physics (INFN), Florence Section, Sesto Fiorentino, Italy

* Email: thomas.kreuz@cnr.it

Introduction
In Kreuz et al., J Neurosci Methods 381, 109703 (2022)[1]two methods were proposed that perform latency correction, i.e., optimise the spike time alignment of sparse neuronal spike trains with well-defined global spiking events. The first one based on direct shifts is fast but uses only partial latency information, while the other one makes use of the full information but relies on the computationally costly simulated annealing. Both methods reach their limits and can become unreliable when successive global events are not sufficiently separated or even overlap.

Methods
Here[2]we propose an iterative scheme that combines the advantages of the two original methods by using in each step as much of the latency information as possible and by employing a very fast extrapolation direct shift method instead of the much slower simulated annealing.

Results
We illustrate the effectiveness and the improved performance, measured in terms of the relative shift error, of the new iterative scheme not only on simulated data with known ground truths but also on single-unit recordings from two medial superior olive neurons of a gerbil. The iterative scheme outperforms the existing approaches on both the simulated and the experimental data. Due to its low computational demands, and in contrast to simulated annealing, it can also be applied to very large datasets.

Discussion
The new method generalises and improves on the original method both in terms of accuracy and speed. Importantly, it is the only method that allows to disentangle global events with overlap.

Acknowledgements
J.M.H. and B.B. were supported in this study by an Australian Research Council Laureate Fellowship (FL 160100108) awarded to D.M.
References
[1]

Kreuz, T., Senocrate, F., Cecchini, G., Checcucci, C., Mascaro, A.L.A., Conti, E., Scaglione, A. and Pavone, F.S., 2022
Latency correction in sparse neuronal spike trains
J. Neurosci. Methods 381, 109703 (2022)
http://dx.doi.org/10.1016/j.jneumeth.2022.109703

[2]

Mariani, A., Senocrate, F., Mikiel-Hunter, J., McAlpine, D., Beiderbeck, B., Pecka, M., Lin, K. and Kreuz, T., 2025
Latency correction in sparse neuronal spike trains with overlapping global events
Journal of Neuroscience Methods 110378
https://doi.org/10.1016/j.jneumeth.2025.110378

Speakers

Thomas Kreuz

Arturo Mariani

Federico Senocrate

Jason Mikiel-Hunter

David McAlpine

Barbara Beiderbeck

Michael Pecka

Kevin Lin

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P148: ELiSe: Efficient Learning of Sequences in Structured Recurrent Networks

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P148 ELiSe: Efficient Learning of Sequences in Structured Recurrent Networks

Laura Kriener¹⸴²
Ben von Hünerbein*¹
Kristin Völk³
Timo Gierlich¹
Federico Benitez¹
Walter Senn¹
Mihai A. Petrovici¹

¹ Department of Physiology, University of Bern, 3012 Bern, Switzerland.
² Institute of Neuroinformatics, University of Zurich and ETH Zurich, Zurich, Switzerland
³ Catlab Engineering GmbH, Grafrath, Germany

*Email: ben.vonhuenerbein@unibe.ch

Introduction
To learn complex action sequences, neural networks must maintain memories of past states. Typically, the required transients are produced by strong network recurrence. The biological plausibility of existing solutions for recurrent weight learning suffers from issues with locality (BPTT [1]), resource scaling (RTRL [2]), or parameter scales (FORCE [3]). To alleviate these, we introduce dendritic computation and a static structural scaffold to our recurrent networks. Leveraging this, our always-on local plasticity rule carves out strong attractors which generate the target activation sequences. We show that with few neurons, our model learns to reproduce complex non-Markovian sequences robustly despite external disturbances.
Methods
Our network contains two populations of structured neurons with somatic and dendritic compartments and leaky-integrator dynamics that integrate presynaptic inputs (Fig. 1a1). Output rates are computed as non-linear functions on the voltage. During development, a sparse scaffold of static somato-somatic connections with random delays is formed (Fig. 1a2,3). A teacher nudges output neurons towards a target pattern, and the somato-somatic scaffold transports this signal throughout the network. The dense, plastic, and randomly delayed somato-dendritic weights (Fig. 1a4)use these signals to adapt based on a local error-correcting learning rule [4].This gives rise to a robust dynamical attractor which generates the correct output pattern in the absence of a teacher.
Results
We demonstrate our model's ability to learn complex, non-Markovian sequences by exposing it repeatedly to a sample of Beethoven's "Für Elise" (Fig. 1b). We find that learning the recurrent weights is critical by showing that our model outperforms a same-size reservoir, both in its ability to learn and then sustain a pattern during replay (Fig. 1c). Next, we demonstrate robust learning across large ranges of the network parameter space. Further, despite severe temporary disruptions of the output population activity during pattern replay, the network is able to recover a correct replay of the learned pattern. Finally, we show that our network is able to extract the denoised signal from noisy target activities.
Discussion
Compared to other models of sequence learning in cortex, we suggest that ours is more resource-efficient, more biologically plausible, and, in general, more robust. It starts with only a sparse, random connection scaffold generating weak and unstructured activity. We show that this is enough for local plasticity to extract useful information in order to imprint strong attractor dynamics, in a manner that is robust to parameter variability and external disturbance. Unlike other approaches, learning in our networks is phaseless and is not switched off during validation and replay.

Figure 1. (a) Development and learning in ELiSe. (a1) Sparse somato-somatic scaffold based on p and q (a2) with interneuron driven inhibition (a3). Dense somato-dendritic synapses (green) adapted during learning (a4). (b) Learning in early, intermediate and final stages (teacher removal at red line). (c) Learning accuracy and stability during learning and replay compared to an equivalent reservoir.
Acknowledgements
We thank Richard Hahnloser and his lab for valuable feedback on learning in songbirds. We gratefully acknowledge funding from the European Union for the Human Brain Project (grant #945539) and Fenix Infrastructure resources (grant #800858), the Swiss National Science Foundation (grants #310030L\_156863 and #CRSII5\_180316) and the Manfred Stärk Foundation.

References
[1] Werbos,Paul J. "Backpropagation through time: what it does and how to do it" Proceedings of the IEEE 78.10 (1990): 1550-1560.
[2] Marschall,Owen,Kyunghyun Cho, and Cristina Savin. "A unified framework of online learning algorithms for training recurrent neural networks." Journal of machine learning research 21.135 (2020): 1-34.
[3]Sussillo,David, and Larry F. Abbott. "Generating coherent patterns of activity from chaotic neural networks" Neuron 63.4 (2009): 544-557.
[4] Urbanczik, Robert, and Walter Senn. "Learning by the dendritic prediction of somatic spiking" Neuron 81.3 (2014): 521-528.

Speakers

Mihai A. Petrovici

Walter Senn

Federico Benitez

Laura Kriener

Ben von Hünerbein

Kristin Völk

Arno Granier

Timo Gierlich

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P149: Ion Channel Contributions to Spike Timing Precision in Computational Models of CA1 Pyramidal Neurons: Implications for Channelopathies

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P149 Ion Channel Contributions to Spike Timing Precision in Computational Models of CA1 Pyramidal Neurons: Implications for Channelopathies

Anal Kumar*1, Upinder S. Bhalla1

1National Centre for Biological Science, Tata Institute of Fundamental Research, Bangalore, India

*Email:analkumar@ncbs.res.in
Introduction

Precise neuronal spike timing is essential for encoding [1,2], phase coding [3,4], and spike-timing-dependent plasticity (STDP) [5,6]. Disruptions in spike timing precision (SpTP) are linked to disorders such as auditory processing disorder [7] and autism spectrum disorder (ASD) [8,9]. These conditions are also associated with channelopathies [8], yet the specific contributions of different ion channels to SpTP remain unclear. In this study, we use computational models of CA1 pyramidal neurons to systematically examine how ion channel overexpression and underexpression affect SpTP, providing insights into disease mechanisms and potential therapeutic targets.

Methods
We constructed data-driven, conductance-based models of CA1 pyramidal neurons, incorporating realistic electrotonic, passive, and active features based on experimental recordings. Twelve ion channel subtypes were included, with kinetics derived from prior studies. To evaluate SpTP, we analyzed the coefficient of variation of inter-spike intervals and jitter slope across multiple trials of tonic 150 pA current injections. Gaussian noise was added to these current injections to simulate physiological noise. To determine the impact of early vs late activating ion channels on SpTP, we assessed SpTP separately for initial and later spikes in the spike train.

Results
Due to heterogeneity in the Gbar of ion channels across models, individual models exhibited variable effects of Gbar on SpTP. However, some global trends emerged:
● Initial spikes in the action potential train: SpTP negatively correlated with HCN and persistent sodium (Na_P) channels, while Kv3.1 showed a positive correlation. Transient sodium (Na_T) channels exhibited a non-monotonic relationship.
● Later spikes in the action potential train: SpTP negatively correlated with Na_P, whereas Kv3.1, K_SK, K_BK, and K_P showed a positive correlation.

Other channels, including K_P, K_T, K_M, K_D, and calcium channels (LVA, HVA), showed no significant impact on SpTP across trials.

Discussion

Previous studies have reported increased K_SK currents and reduced SpTP of later spikes in Fragile X Syndrome (FXS) [8]. Our findings corroborate this by demonstrating a positive correlation between K_SK Gbar and SpTP of later spikes, suggesting that K_SK upregulation may contribute to impaired temporal precision in FXS. Additionally, our study identifies potential therapeutic targets, such as Na_P channel blockade, which may help counteract the SpTP deficits observed in FXS. Further analysis of these models will help uncover the underlying mechanisms driving these correlations, shedding light on the role of ion channel dysfunction in neurodevelopmental disorders.

Acknowledgements
We thank NCBS, TIFR and Department of Atomic Energy, Government of India, under project identification No. RTI 4006 for funding. Special thanks to Dr. Deepanjali Dwivedi and Anzal KS for the raw experimental recordings. Thanks to NCBS animal house, Imaging facility, super computing facility at NCBS and members of Bhalla Lab.
References
● https://doi.org/10.1126/science.1149639
● https://doi.org/10.1103/PhysRevLett.80.197
● https://doi.org/10.1038/nature02058
● https://doi.org/10.1002/hipo.450030307
● https://doi.org/10.1523/JNEUROSCI.18-24-10464.1998
● https://doi.org/10.1126/science.275.5297.213
● https://doi.org/10.1016/j.heares.2015.06.014
● https://doi.org/10.1523/ENEURO.0217-19.2019
● https://doi.org/10.1016/j.neuron.2017.12.043

Speakers

Upinder Bhalla

Anal Kumar

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P150: Network vs ROI Perspectives: Brain Connectivity Analysis using Complex Principal Component Analysis

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P150 Network vs ROI Perspectives: Brain Connectivity Analysis using Complex Principal Component Analysis

Puneet Kumar*1†, Alakhsimar Singh2†, Xiaobai Li1,3, Shella Keilholz4, Eric H. Schumacher5

1University of Oulu, Finland
2National Institute of Technology Jalandhar, India
3Zhejiang University, China
4Emory University, USA
5Georgia Institute of Technology, USA

†Equal Contribution
*Email: puneet.kumar@oulu.fi

Introduction: We implement Complex Principal Component Analysis (CPCA) [1] for brain connectivity analysis. It largely reproduces traditional Quasi-Periodic Patterns (QPP)-like activity [2] and handles tasks of various lengths, while QPP struggles with shorter tasks. We present network- and ROI-level observations for the Human Connectome Project (HCP) data having four 15-min rest scans (TR=0.72s) and seven tasks (1 hour total) [3]. Our focus is on Task-Positive Network (TPN) – defined as the Dorsal Attention Network (DAN) plus Fronto-Parietal Network (FPN), and Default Mode Network (DMN). Our contributions are CPCA implementation and dual (network and ROI) level analysis. The implementation code is at github.com/MIntelligence-Group/DBCATS.
Methods: The data was preprocessed using the Configurable Pipeline for the Analysis of Connectomes (C-PAC) [4], including motion and slice-timing correction, normalization to MNI space, and band-pass filtering (0.01–0.1 Hz). We focus on the working memory (0-back/2-back) task with 405 frames/run. Each run has eight 42.5 s task blocks (10 trials of 2.5 s), four 15 s fixation blocks, and 2 s stimuli followed by a 500 ms ITI. The DMN (36 ROIs), DAN (33 ROIs), and FPN (30 ROIs) were defined using the 7-network parcellation [5]. We adapted CPCA for fMRI by applying the Hilbert transform to introduce a 90° phase shift, capturing amplitude and phase. Seven principal components (PCs) were extracted to reconstruct the dominant activity patterns.
Results: Fig. 1(a) and 1(e) display Blood Oxygenation Level Dependent (BOLD) activation at the global network level for rest and task states. Correlation values between DMN and DAN are -0.99, and between DMN and FPN -0.91, as per Fig. 1(i) and 1(j). Fig. 1(b–d) depicts local ROI-level BOLD activation (from both left and right hemispheres of the brain) during rest, and Fig. 1(f–h) during task. In the rest state, FPN shows 440 positive and 618 negative correlations with DMN, and DAN shows 531 and 629. For the task state, FPN has 439 positive and 620 negative correlations with DMN, and DAN has 532 and 629. Comparing Fig. 1(k–n) indicates slightly shifted connectivity patterns from rest to task, reflecting changes in DMN, DAN, and FPN signals.
Discussion: At the network level, DMN shows anticorrelation with both DAN (-0.99) and FPN (-0.91), as depicted in Fig. 1(i,j). At the ROI level, 44% (1972) of DMN-TPN pairs are positively correlated, while 56% (2496) are negative, indicating local differences. Correlations become more negative from rest to task, though changes are modest. Fig. 1(k–n) shows these changes, highlighting how brain connections adapt at the ROI level and exhibit task-dependent shifts. We are the first to implement CPCA as a potential brain connectivity analysis method comparing rest and task. We aim to extend our implementation to other datasets, seeking visibility for our work and findings and feedback to refine our approach and drive further advancements.

Figure 1. Network-level and ROI-level BOLD time series for DMN, DAN, and FPN during rest (a–d) and task (e–h). Network-level correlation connectivity matrices (CCM) (i, j). ROI-level CCMs for DMN–DAN regions (k, l) and DMN-FPN regions (m, n) for rest and task. (a, e) show average PC1 activity at network level, while (b–d, f–h) show PC1 activity at ROI level, with different colors denoting different ROIs.
Acknowledgements
The authors gratefully acknowledge the collaboration with the CoNTRoL Lab and GSU/GT Center for Advanced Brain Imaging at Georgia Institute of Technology, USA, and the Keilholz Mind Lab at Emory University, USA. We thank the CMVS International Research Visit Program 2024 for funding and the University of Oulu, Eudaimonia Institute, and CSC Finland for support and computational resources.
References
[1] Bolt, T.,... (2022). A Parsimonious Description of Global Functional Brain Organization in Three Spatiotemporal Patterns. Nature Neuroscience, 25(8), 1093-1103.
[2] Abbas, A.,... (2019). Quasi-Periodic Patterns Contribute to Brain Functional Connectivity. Neuroimage, 191, 193-204.
[3] Van Essen, D. C.,... (2012). The Human Connectome Project. Neuroimage, 62(4), 2222-2231.
[4] Craddock, C.,... (2013). Towards Automated Analysis of Connectomes. Front Neuroinform, 42 (10).
[5] Yeo, B. T.,... & Buckner, R. L. (2011). The Organization of Human Cerebral Cortex. Journal of Neurophysiology.

Speakers

Puneet Kumar

Alakhsimar Singh

Xiaobai Li

Shella Keilholz

Eric H. Schumacher

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P151: Between near and far-fields: influence of neuronal morphology and channel density on EEG-like signals

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P151 Between near and far-fields: influence of neuronal morphology and channel density on EEG-like signals

Paula T. Kuokkanen*1, Richard Kempter1,2,3, Catherine E. Carr4, Christine Köppl5

1Institute for Theoretical Biology, Humboldt-Universität zu Berlin, 10115 Berlin, Germany
2Bernstein Center for Computational Neuroscience Berlin, 10115 Berlin, Germany
3Einstein Center for Neurosciences Berlin, 10115 Berlin, Germany
4Department of Biology, University of Maryland College Park, College Park, MD 20742
5Department of Neuroscience, School of Medicine and Health Sciences, Research Center for Neurosensory Sciences and Cluster of Excellence “Hearing4all” Carl von Ossietzky University, 26129 Oldenburg, Germany

*Email: paula.kuokkanen@hu-berlin.de
Introduction

Both the near and far fields of extracellular neural recordings are well understood. The near field can be explained by models of ion channel activity in nearby compartments [1]. The far field can be approximated by current dipoles produced by the membrane currents of multicompartmental cells [2]. The dipole spanned between the dendrites and soma is typically assumed to be the basis of the electro-encephalography (EEG) signals of cortical pyramidal neurons [e.g. 3]; yet also their somatic spikes can be observed in the EEG [4]. Such potentials, measured relatively far away from the source but not strictly in the far field, are highly dependent on the morphology of the cell, its ion channel concentrations, and the electrodes’ positions [5].

Methods
We simulate single multi-compartment cells with NEURON and LFPy packages to study their 'mid-field' potentials. We vary the neurons’ simplified morphologies systematically, and use combinations of channel densities to compare the mid-field potentials with the dipole moments of the cells. We especially study the spatial limitation of the far-field approximation depending on the cell properties. We verify our results with the use of experimental data [6], EEG-like single-cell recordings from the auditory nerve and the auditory brainstem Nucleus Magnocellularis in the barn owl.

Results
We observe that, as expected, the dendritic-somatic dipole can determine the far and mid-fields in pyramidal cell-like morphologies. Unexpectedly, we observe that a dipole moment caused by branching axons can have a similar amplitude to the dendritic dipole in mid and far fields. Furthermore, we show that under certain conditions a somatic spike — not necessarily related to any current dipole — can contribute to fields even at a distance of 10 mm from the soma. These results match with our experimental results from the owl.
Discussion
Common assumptions about the distances from a neuronal source where far-field conditions predominate may not hold. Depending on the neuron type, both the morphology and differential densities of active ion channels across cell compartments can play a large role in creating their fields at varying distances. The axonal arborizations, because activated simultaneously by a single spike, can create a dipole [7] with a surprisingly large contribution to the far fields as compared to the dendritic-somatic dipoles. Furthermore, large somata with high densities of active currents can contribute to the extracellular field at distances of even 1 cm, violating the usual far-field assumption.

Acknowledgements
We thank Ghadi ElHasbani for helpful discussions, and Hannah Schultheiss for preliminary modeling.
This research was funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) grant nr. 502188599.
References
1. https://doi.org/10.1152/jn.00979.2005
2. https://doi.org/10.1016/j.neuroimage.2020.117467
3. https://doi.org/10.7554/eLife.51214
4. https://doi.org/10.1016/j.neuroimage.2014.12.057
5. http://doi.org/10.1097/00004691-199709000-00009
6. https://doi.org/10.1101/2024.05.29.596509
7. https://doi.org/10.7554/eLife.26106

Speakers

Paula Kuokkanen

Catherine Carr

Christine Köppl

Richard Kempter

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P152: Differing Strategies and Neural Representations in the Same Long-Term Information Encoding Task

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P152 Differing Strategies and Neural Representations in the Same Long-Term Information Encoding Task

Tomoki Kurikawa*1

1Department of Complex and Intelligence Systems, Future University Hakodate, Hakodate, Japan
*Email: kurikawa@fun.ac.jp

Introduction

Many cognitive tasks require maintaining information across trials, such as deterministic or probabilistic reversal learning tasks. In the deterministic reversal learning task [1], for instance, the pairing between sensory cues and behavioral outcomes reverses after a fixed number of trials. To perform such a task successfully, subjects have to track the number of elapsed trials to predict reversals accurately. However, the neural representation underlying such sustained memory processes remains poorly understood.

Methods
To uncover the representations underlying task performance, we built a simple recurrent neural network (RNN) model trained on a deterministic reversal learning task using machine learning techniques. We analyzed what representations emerged and how they were formed. In this task, there were two types of blocks, and depending on the block type, the network had to alternate between two outputs (Left and Right outputs). Each block consisted of 10 trials, and the block type switched every 10 blocks iteratively. Notably, no explicit contextual cues were provided—the network had to track trial counts internally. The model was trained to produce correct outputs across 10 consecutive blocks.

Results
We found that two distinct strategies emerged after learning 10 blocks: generalization and specification. In the generalization strategy, the network discovered the underlying rule of the task. Despite being trained on only 10 blocks, it could generalize and perform correctly beyond this limit. In contrast, in the specification strategy, the network was specifically trained to complete the 10-block task but was unable to extend its performance to a larger number of blocks, such as a 20-block task.
What representations underlie these different behaviors? Our analysis revealed that different neural representations support these distinct strategies. In the generalization strategy, certain neurons specifically encoded the number of trials within a block. Their activity gradually increased across trials, and when a threshold was reached, the network switched outputs from one output to another before resetting, indicating that these neurons tracked the number of trials within a block.
In contrast, in the specification strategy, no individual neurons encoded trial counts explicitly. Instead, this information was distributed across the neural population, implying a different mechanism for task execution.

Discussion
Our findings suggest that even when performing the same task, different strategies can emerge across subjects or animals. Depending on the adopted strategy, the way long-term information is encoded across trials also varies. This computational result provides new insights into how long-term information is represented in neural systems.

Acknowledgements
The present work is supported bySpecialResearch Expenses in Future University Hakodate
References
https://doi.org/10.1523/ENEURO.0172-24.2024

Speakers

Tomoki Kurikawa

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P153: Efficient estimation of mutual-information rates from spiking data by maximum-entropy models

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P153 Efficient estimation of mutual-information rates from spiking data by maximum-entropy models

Tobias Kühn*1, Gabriel Mahuas1, Ulisse Ferrari1

1Institut de la Vision, Sorbonne Université, CNRS, INSERM, Paris, France

*Email: tkuehn@posteo.de
Introduction

Neurons in sensory systems encode stimulus information into their stochastic spiking response. This is quantified by the mutual-information rate (MIR), given by the ratio of the mutual information between the activity of a spiking neuron and a (dynamical) stimulus and time. The computation of the MIR is challenging, because it requires the estimation of entropies, in particular the ones conditional on the stimulus. However, this is difficult in the realm of correlated, poorly sampled data, for which estimates are prone to biases.
Methods
We here present moment-based the mutual-information-rate approximation (Moba-MIRA), a computational method to estimate the MIR. It is based on the idea of taking into account the statistics of the activity single time bins exactly and consider the correlations of the activity between them by employing a statistical model featuring pairwise interactions, similar to the Ising model of statistical physics. This is similar to other maximum-entropy approaches employed in neuroscience, however, we do not restrict our spike counts to be binary, allowing the use of relatively large time bins. To achieve the estimate of the entropies, we use a (Feynman) diagrammatic expansion in the covariances between the activities of all time bins [1,2,3].

ResultsWe test our method on artificial data from a generalized linear model mimicking the activity of retinal ganglion cells and demonstrate that we approximate the exact result in the well-sampled regime in a satisfactory way. Importantly, our method introduces only a limited bias even in case of a number of samples attainable in experiments, about 60 to 100, allowing it to use it for to real data. Applying it to ex-vivo electrophysiological recordings from rat retinal-ganglion cells (on and off), stimulated by black-and-white checkerboards or bars moving in a random way, we obtain information rates of about 2 to 20 bits/s for every neuron, consistent with values from the literature.

DiscussionTested on artificial data, Moba-MIRA outperforms the state-of-the-art method [4] - depending on the variant clearly in speed, with comparable precision, or in precision, with comparable speed, compare figure. We therefore believe that it can serve as a efficient and simple tool for the analysis of spiking data. In particular, extending it to be applicable to populations of neurons is easy, so that it will allow the study of collective effects in addition to the effects coming about by neuronal dynamics.

Figure 1. a) Estimate of the MIR for artificial, retina-like data with state-of-the-art method by Strong et al. (histogram) and our approach. In the latter, we estimate the entropy conditional on the stimulus by a maximum-entropy model, for which we show the compute time in panel b.
Acknowledgements
We acknowledge ANR for financial support.
References
[1] Tobias Kühn and Moritz Helias. Expansion of the effective action around non-gaussian theories. Journal of Physics A: Mathematical and Theoretical, 51(37):375004, Aug 2018.
[2] Tobias Kühn and Frédéric van Wijland. Diagrammatics for the inverse problem in spin systems and simple liquids. Journal of Physics A: Mathematical and Theoretical, 56(11):115001, Feb 2023.
[3] Gabriel Mahuas, Olivier Marre, Thierry Mora, and Ulisse Ferrari. Small-correlation expansion to quantify information in noisy sensory systems. Phys. Rev. E, 108:024406, Aug 2023.

Speakers

Tobias Kühn

Gabriel Mahuas

Ulisse Ferrari

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P154: Comparison of derivative-based and correlation-based methods to estimate effective connectivity in neural networks

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P154 Comparison of derivative-based and correlation-based methods to estimate effective connectivity in neural networks

Niklas Laasch1, Wilhelm Braun1,2, Lisa Knoff1, Jan Bielecki2, Claus C. Hilgetag1,3

1Institute of Computational Neuroscience, Center for Experimental Medicine, University Medical Center Hamburg-Eppendorf, Martinistraße 52, 20246, Hamburg, Germany

2Faculty of Engineering, Department of Electrical and Information Engineering,Kiel University, Kaiserstrasse 2, 24143, Kiel, Germany

3Department of Health Sciences, Boston University, 635 Commonwealth Avenue, Boston, MA, 02215, USA

E-Mail:niklas.laasch@posteo.de
Introduction
Inferring effective connectivity in neural systems from observed activity patterns remains a challenge in neuroscience. Despite numerous techniques being developed, no universally accepted method exists for determining how network nodes mechanistically affect one another. This limits our understanding of neural network structure and function. We focus on purely excitatory networks of small to intermediate size with continuous dynamics to systematically compare different connectivity estimation approaches, aiming to identify the most reliable methods for specific network characteristics.Methods
We used the Hopf neuron model with known ground truth structural connectivity to generate synthetic neural activity data. Multiple connectivity inference algorithms were applied to reconstruct the system's connectivity matrix, including lagged cross-correlation (LCC) [1], derivative-based covariance analysis (DDC) [2], and transfer entropy methods. We varied parameters controlling bifurcation, noise, and delay distribution to test method robustness. Forward simulations using estimated connectivity matrices were performed to evaluate each method's ability to recreate observed activity patterns. Finally, we applied promising methods to empirical data fromC. elegans.Results
In sparse non-linear networks with delays, combining LCC with DDC analysis provided the most reliable connectivity estimation. LCC performed comparably to transfer entropy in linear networks but at significantly lower computational cost. Performance was optimal in small sparse networks and decreased in larger, denser configurations. With the Hopf model, LCC-based connectivity estimates yielded higher trace-to-trace correlations than derivative-based methods for sparse noise-driven systems. When applied toC. elegansneural data, LCC outperformed more computationally expensive methods, including a reservoir computing approach.Discussion

Our findings demonstrate that a comparatively simple method - lagged cross-correlation - can reliably estimate directed effective connectivity in sparse neural systems despite spatio-temporal delays and noise. This has significant implications for biological research scenarios where only neuronal activity, but not connectivity or single-neuron dynamics, is observable. We provide concrete suggestions for effective connectivity estimation in such common research scenarios. Our work contributes to bridging the gap between observed neural activity and underlying network structure in neuroscience.

Acknowledgements
The authors would like to thank Kayson Fakhar, Alexander Schaum, Fatemeh Hadaeghi, Arnaud Messé, Gorka Zamora-López and Heike Siebert for useful comments.
References
[1] 10.1038/s41598-025-88596-y
[2]10.1073/pnas.2117234119

Speakers

Niklas Laasch

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P155: Bayesian Modelling of Explicit and Implicit Timing

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P155 Bayesian Modelling of Explicit and Implicit Timing

Gianvito Laera*1,2,3,4, Matthew Vowels5,6,7, Tasnim Daoudi1,2, Richard Andrè1,2, Sam Gilbert8, Sascha Zuber2,3, Matthias Kliegel1,2,3, Chiara Scarampi2,3

1Cognitive Aging Lab (CAL), Faculty of Psychology and Educational Sciences, University of Geneva, Switzerland
2Centre for the Interdisciplinary Study of Gerontology and Vulnerability, University of Geneva, Switzerland
3LIVES, Overcoming Vulnerability: Life Course Perspective, Swiss National Centre of Competence in Research, Switzerland
4University of Applied Sciences and Arts Western Switzerland HES-SO, Geneva School of Health Sciences, Geneva Musical Minds lab (GEMMI lab), Geneva, Switzerland
5Institute of Psychology, University of Lausanne, Switzerland
6The Sense Innovation and Research Center, CHUV, Switzerland
7Centre for Vision, Speech and Signal Processing, University of Surrey, Switzerland
8Institute of Cognitive Neuroscience, University College London, London, United Kingdom

*Email: gianvito.laera@unige.ch

Introduction
Time perception supports adaptive behavior by allowing anticipation of critical events [1]. Explicit timing involves conscious estimation of durations (e.g., interval reproduction), typically modeled by Bayesian frameworks combining noisy sensory evidence with prior expectations [2]. Implicit timing emerges indirectly through tasks like foreperiod paradigms, relying on neural or motor strategies without temporal awareness. Historically treated separately, explicit tasks engage cortico-striatal circuits, whereas implicit tasks involve cerebellar or parietal regions. We hypothesized that a unified Bayesian model with a shared internal clock parameter (θ) could bridge explicit and implicit timing abilities.
Methods
Forty-five psychology students performed four within-participant tasks: Explicit Motor (spontaneous motor response), Implicit Motor (simple reaction time), Explicit Temporal (interval reproduction), and Implicit Temporal (stimulus prediction). A hierarchical Bayesian model estimated an internal clock rate parameter (θ), reflecting subjective timing (θ=1 accurate; θ>1 slower; θ<1 faster clock), alongside parameters modeling task-specific variability and individual learning effects. Explicit tasks involved duration reproduction without feedback; implicit tasks involved temporal anticipation of a stimulus. Markov Chain Monte Carlo (MCMC) sampling via Stan was used for parameter estimation.
Results
The Bayesian model indicated participants’ internal clocks ran faster than objective time (μθ≈0.80), explaining interval overestimation at short durations and confirming a regression-to-mean effect. Individual differences in θ were significant (τθ≈0.20); participants with fewer practice trials had internal clocks closer to accuracy, indicating efficient learning. Explicit tasks had higher variability than implicit tasks, confirming greater cognitive uncertainty. Implicit tasks showed typical foreperiod effects (longer expected intervals slightly slowed reaction times,a≈0.3). Explicit and implicit timing shared moderate variance (r≈0.45), and network analysis suggested θ centrally bridged both timing domains.
Discussion
The findings support a unified Bayesian model, highlighting a shared internal clock mechanism underlying explicit and implicit timing. The internal clock parameter (θ) explained significant individual differences across tasks, supporting recent integrative views proposing partially overlapping neural substrates [3, 4]: a common cognitive mechanisms (maybe striatal-thalamo-cortical circuits) provide duration information utilized differently across explicit versus implicit tasks. Task-specific differences also comprise additional factors (e.g., cognitive strategies, attention and memory load) that future versions of the model should include. The model can be promising in explaining timing difficulties in clinical and aging populations too.

AcknowledgementsNone
References
1. https://doi.org/10.1016/j.neuropsychologia.2012.08.017
2. https://doi.org/10.1016/j.tics.2013.09.009
3. https://doi.org/10.1016/j.cobeha.2016.01.004
4. https://doi.org/10.1016/j.tins.2004.10.007

Speakers

Gianvito Laera

Matthew Vowel

Tasnim Daoudi

Richard Andrè

Sam Gilbert

Zuber Sascha

Matthias Kliegel

Chiara Scarampi

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P156: Population decoding of visual motion direction in V1 marmoset monkey : effects of uncertainty

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P156 Population decoding of visual motion direction in V1 marmoset monkey : effects of uncertainty

Alexandre C. Lainé*1, Sophie Denève1, Nicholas J. Priebe2, Guillaume S. Masson1, Laurent U. Perrinet1

1Institut de Neurosciences de la Timone, UMR 7289, CNRS - Aix-Marseille University, Marseille, France.
2Section of Neurobiology, School of Biological Sciences, University of Texas at Austin, Austin, TX, USA.

*Email: alexandre.laine@univ-amu.fr
Introduction

Studying the internal representation of information in the primary visual cortex (V1) is crucial to understand how we perceive the external world. Research on 2D motion direction in non-human primates [1,2,3] in particular when displaying naturalistic stimuli like MotionClouds [4] reveals substantial diversity and multiple mechanisms within the neuronal population [5]. This project aims to examine how a large population of V1 neurons encodes stimulus direction by explicitly titrating the precision in the orientation and spatial frequency domains.

Methods
Activity of several hundreds of neurons was recorded using Neuropixel 2.0 technology [6] in area V1 of an anesthetized marmoset monkey during which MotionClouds were presented for eight directions and two precision levels. We use a decoding method to analyze the representation of motion direction in the marmoset V1, focusing on the effects of uncertainty on the population code. The decoding method optimizes the weights of a logistic regression to achieve optimal decoding accuracy on a training set. Training can be conducted (1) on a broad time window, (2) by applying temporal generalization [7], or (3) after reducing dimensionality with dPCA [8].
Results
After training on broad windows, analysis of the optimised weights revealed two types of population representations: transient and sustained. These representations differ in their distributions across cortical layers, confirming earlier results obtained in another species [5], and are modulated by the level of orientation precision. The accuracy measured on the test set revealed first that a broad spatial frequency distribution leads to a better decoding performance, and second that the precision of the orientation is a critical factor in the representation of motion direction. Indeed, a high precision in orientation leads to the aperture problem, and thus to an ambiguity in the representation of motion direction. Temporal generalization confirm a stable representation. Projection of neuronal activity using dPCA onto 10 components without affect accuracy demonstrate that the information may be represented in a low-dimensional manifold.

Discussion
In summary, this decoding method clarifies how directional information is represented and modulated by precision in marmoset V1. The coexistence of transient and sustained representations indicates distinct functional roles across cortical layers. Temporal generalization confirms that the neuronal population maintains a stable encoding of direction. Reducing dimensionality while preserving precision implies that a small set of components can capture the essential features of neuronal activity, enabling the exploration of various projection methods to optimize decoding. Moreover, the results suggest that orientation precision could be a major factor in shaping the interplay between orientation and direction.

Acknowledgements
This work was supported by ANR-NSF CRCNS grant “PrioSens” N° ANR-20-NEUC-0002 attributed to G.S.M,
N.J.P. and L.U.P. and by a doctoral grant from the French Ministry of Higher Education and Research, awarded
by the Doctoral School 62 of Aix-Marseille University to A.L.
References[1] https://doi.org/10.1113/jphysiol.1959.sp006308
[2] https://doi.org/10.1113/jphysiol.1968.sp008455
[3] https://doi.org/10.1523/JNEUROSCI.1335-12.2012
[4] https://doi.org/10.1152/jn.00737.2011
[5] https://doi.org/10.1038/s42003-023-05042-3
[6] https://doi.org/10.1126/science.abf4588
[7] https://doi.org/10.1016/j.tics.2014.01.002
[8] https://doi.org/10.7554/eLife.10989

Speakers

Alexandre Lainé

Nicholas J. Priebe

Guillaume S. Masson

Laurent U. Perrinet

Sophie Deneve

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P157: Non-monotonic Subthreshold Information Filtering in a Coupled Resonator-Integrator System

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P157 Non-monotonic Subthreshold Information Filtering in a Coupled Resonator-Integrator System

Franquelin Lambert1

1Université de Moncton, Département de physique et d'astronomie
Introduction:Subthreshold dynamics play a key role in spike generation, and it is well-known that some neurons exhibit a frequency preference when integrating subthreshold input– so-called resonators [1,2]. It has been shown, however, that despite the existence of subthreshold resonance, a single resonator neuron exhibits low-pass, i.e., monotonic, information filtering(as measured by the spectral coherence). In other words, in the subthreshold regime, band-pass impedance does not translate to band-pass information filtering. Instead, nonlinearities, such as spiking dynamics, are needed to create band-pass information transfer [3,4].

Methods:Here, we study a similar question in an electrically coupled pair of neurons. Our goal is to evaluate whether this resonance profile imparts non-trivial information filtering capabilities to the coupled system. We numerically simulate an electrically coupled integrate-and-fire to resonate-and-fire system in the subthreshold regime, and we investigate the stimulus-response spectral coherence function of the system under perturbation by coloured noise (Ornstein-Uhlenbeck process).

Results:For electrical coupling between a resonator and an integrator, we show that a Fano-like resonance profile appears in the impedance, i.e., a narrow, asymmetric peak with anti-resonance [5]. Moreover, we observe that the coherence function is non-monotonic, with a minimum around the frequency of the opposite neuron.

Discussion:This challenges the claim that neurons require nonlinearities to relay bandpass information filtering properties. This new perspective places informationfiltering in the context of connection motifs where a small number of resonators and integrators interact, rather than the context of individual neurons.

Acknowledgements
no acknowledgements
References
[1] Izhikevich, Eugene M.Dynamical systems in neuroscience. MIT press, 2007.
[2]https://doi.org/10.1016/S0893-6080(01)00078-8
[3]https://doi.org/10.1109/TMBMC.2016.2618863
[4]https://doi.org/10.1007/s10827-015-0580-6

[5]https://doi.org/10.1088/0031-8949/74/2/020

Speakers

Franquelin Lambert

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P158: A unified model for estimating short- and long-term synaptic plasticity from stimulation-induced spiking activity

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P158 A unified model for estimating short- and long-term synaptic plasticity from stimulation-induced spiking activity

Arash Rezaei1,2, Mojtaba Madadi Asl3,4,Milad Lankarany*1,2,5

1Krembil Brain Institute, University Health Network, Toronto, ON, Canada
2Institute of Biomedical Engineering, University of Toronto, Toronto, ON, Canada
3School of Biological Sciences‎,‎Institute for Research in Fundamental Sciences (IPM)‎,‎Tehran‎,‎Iran
4Pasargad Institute for Advanced Innovative Solutions (PIAIS)‎,‎Tehran‎,‎Iran
5Center for Advancing Neurotechnological Innovation to Application (CRANIA), Toronto, ON, Canada

*Email:milad.lankarany@uhn.ca
Introduction

Abnormal brain activity is the hallmark of several brain disorders such as Parkinson’s disease, essential tremor, and epilepsy [1,2]. Stimulation-induced reshaping of the brain’s networks through neuroplasticity may disrupt neural activity as well as synaptic connectivity and potentially restore healthy brain dynamics. Synaptic plasticity has been the target of invasive therapies, such as deep brain stimulation [3-5]. Mathematical frameworks were able to estimate short-term [6,7] and long-term [8] synaptic dynamics separately. However, the characterization of both short and long-term synaptic plasticity from spiking activity is crucial for understanding the underlying mechanisms and optimization of spatio-temporal patterns of stimulation.

Methods
We developed a novel synapse model to integrate both short- and long-term plasticity into a unified framework wherein the postsynaptic neuron behaves according to both plasticity mechanisms. In the proposed model, the postsynaptic neuron is notably driven by the STP synaptic current and LTP synaptic weight at each step. To induce short- and long-term synaptic responses, presynaptic spike trains were applied for durations of a few hundred milliseconds (STP experiment) and hundreds of seconds (LTP experiment), respectively, to a single postsynaptic neuron. For the STP experiment, a single presynaptic spike train was used, whereas the LTP experiment involved 1000 presynaptic inputs. For both experiments depressing STP synapses were utilized.
Results
Our results demonstrated that in the STP experiment, the unified model produced the same transient fluctuations in the membrane potential of the postsynaptic neuron as observed in the STP-only model. This is evident when comparing Fig. 1.A and B as we see the same pattern of behavior in the postsynaptic membrane potential. In the LTP experiment, we observed similar long-term distribution of the synaptic weights as in the model with only long-term synapses (Fig. 1.C). However, the depression was more pronounced in the unified model due to the concurrent influence of STP and LTP on the postsynaptic neuron. The number of synapses with lower weights increases with the addition of the depressive STP mechanism compared to the LTP-only model.
Discussion
These findings suggest that the integration of STP and STDP within a single synaptic framework can effectively capture both transient and long-lasting plasticity effects. Furthermore, such uniform modeling of STP and LTP enables the incorporation of various combinations of synaptic settings into a population of neurons. This can potentially enhance the biological plausibility and flexibility of the current stimulation-induced neural models.

Figure 1. Fig. 1. Results of the STP and LTP experiments. A) Input spike train, neural and synaptic behavior of a model with only STP after stimulation. B) Behavior of the unified model with both STP and LTP after stimulation. The postsynaptic neuron was stimulated for 1200 ms with a 20 Hz firing rate and a depressing STP synapse (Red lines: postsynaptic membrane potential, Blue dotted lines: STP synaptic c
Acknowledgements
NA
References
1.https://doi.org/10.1016/j.neuron.2006.09.020
2.https://doi.org/10.1371/journal.pcbi.1002124
3.https://doi.org/10.1002/ana.23663
4.https://doi.org/10.1002/mds.25923
5.https://doi.org/10.1016/j.brs.2016.03.014
6.https://doi.org/10.1371/journal.pcbi.1008013
7.https://doi.org/10.1371/journal.pone.0273699
8.https://doi.org/10.1162/neco_a_00883

9.https://doi.org/10.7554/eLife.47314

Speakers

Milad Lankarany

Arash Rezaei

Mojtaba Madadi Asl

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P159: Neun, an efficient and customizable open-source library for computational neural modeling and biohybrid circuit design.

P162: Pavlovian Conditioning of a Superburst Generating Neural Network for High-precision Perception of Spatiotemporal Sensory Information

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P162 Pavlovian Conditioning of a Superburst Generating Neural Network for High-precision Perception of Spatiotemporal Sensory Information

Kyoung J. Lee*¹, Jongmu Kim², Woojun Park¹, Inhoi Jeong¹

¹ Department of Physics, Korea University, Seoul, Korea
² Department of Mechanical Engineering, Korea University, Seoul, Korea

*Email:kyoung@korea.ac.kr
Introduction

How the brain perceives, learns, and distinguishes different spatiotemporal sensory information remains a fundamental yet largely unresolved question in neuroscience [1]. This study demonstrates how an initially random network of Izhikevich neurons can learn, encode, and differentiate time intervals ranging from milliseconds to tens of milliseconds with high temporal precision using a Pavlovian conditioning framework [2]. Notably, our findings highlight the potential role of superbursts in sensory perception, offering new insights into how neural circuits process temporal information.

Methods
Our network model comprises excitatory and inhibitory neurons with synaptic weights evolving through dopamine-modulated spike-timing-dependent plasticity. The conditioning protocol involves sequential electrical stimulation of, for example, three neuron subpopulations (S0, S1, S2) with specific time intervals (Dt1cond.,Dt2cond.), referred to as “target triplet stimulation.” Despite the presence of various distracting stimuli with different time intervals, the network successfully encodes the target stimulation pattern and later responds to it by generating a distinctive population burst—a neuronal spiking avalanche—which acts as a test gauge for perception.

Results
During conditioning, the initially random network evolves into a feedforward structure [3] (Fig. 1A), where three subpopulations (S0, red; S1, blue; S2, green) self-organize according to the imposed time intervals (Dt1cond.,Dt2cond.), effectively encoding temporal information into its network morphology. With axonal conduction delays, the network generates superbursts, featuring multiple sub-burst humps, lasting tens of milliseconds (Fig. 1B). In a perception test, stimuli with varying time intervals and subpopulations produce distinct neuronal avalanches: For example, a network conditioned forDt1cond.=Dt2cond.= 11 ms exhibits systematically varying burst patterns upon receiving different stimulli (Fig. 1B and 1C).

Discussion
These findings provide insight into how seemingly simple neural circuits can encode and process temporal information through structured population spiking activity. Perception in this system can utilize the shape of stimulus-triggered population bursts, allowing for superb temporal resolution (< 1 ms). Furthermore, incorporating axonal conduction delays enables the network to generate superbursts lasting tens of milliseconds, with intricate internal temporal structures, significantly enhancing its perceptual dynamic range. This learning framework can be extended to distinguish much more complex spatiotemporal sequences beyond the simple triplet examples explored in this study.

Figure 1. Fig. 1 Encoding different sets of (Dt_1^cond., Dt_2^cond.) into network morphology (A) and perceptual testing with various (Dt_1^test, Dt_2^test) combinations (B) and subpopulations (C) for the case of (Dt_1^cond.= Dt_2^cond. = 11 ms). In (A), the colored crossbars mark the centroids of S0, S1, and S2 , reflecting the topographic encoding of temporal information (six different cases are shown).
Acknowledgements
This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (No. RS-2024-00335928).
References
1.https://doi.org/10.1016/j.neuron.2020.08.020
2.https://doi.org/10.1093/cercor/bhl152
3.https://doi.org/10.1371/journal.pcsy.0000035

Speakers

Kyoung Jin Lee

In Hoi Jeong

Jongmu Kim

Woojun Park

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P163: Distinct disinhibitory circuits differentially modulate the multisensory integration in a feedforward network

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P163 Distinct disinhibitory circuits differentially modulate the multisensory integration in a feedforward network

Seung-Youn Lee*1,2, Kyujin Kang1,3, Yebeen Yoon1,3, Jae-Ho Han2,3, Hyun Jae Jang1

1Korea Institute of Science and Technology, Seoul, Republic of Korea
2Department of Artificial Intelligence, Korea University, Seoul, Republic of Korea
3Department of Brain and Cognitive Engineering, Korea University, Seoul, Republic of Korea

*Email: seungyounlee@korea.ac.kr

Introduction

Multisensory integration is a fundamental neural process that combines simultaneously presented unisensory inputs into a unified perception. For effective multisensory processing, cross-modal integration via neural network should be dynamically modulated spanning cortical and subcortical regionsin vivo[1,2]. One such mechanism is the disinhibitory circuit gating local information flow by inhibiting other inhibitory neurons. However, it is unclear whether disinhibitory circuits modulate multisensory integration locally or via long-range projections[3,4]. Therefore, we investigated how distinct disinhibition architectures differentially modulate long-range cross-modal integration, such as between the primary auditory cortex (A1) and the visual cortex (V1)[5].

Methods
To test this, we developed a computational feedforward network model incorporatingin vivo-recorded spike trains from A1 and V1. The model consists of two four-layer columns, each representing a different sensory modality, converging onto an output layer (LOUT). Neurons were modeled as single-compartment Hodgkin-Huxley-type neurons, capturing the electrophysiological properties of pyramidal (PYR), somatostatin-positive (SST+), and vasoactive intestinal polypeptide-positive (VIP+) neurons. The disinhibitory circuit was modeled such that VIP+inhibit SST+, which in turn inhibit PYR. The first layer of each column received spike trains recordedin vivoA1 and V1 during pure-tone and grating stimulation as inputs.

Results
We first assessed disinhibitory circuits in multisensory integration. A network with disinhibition exhibited higher mutual information (MIrate) between stimulus variables and firing rates of Loutthan one without, indicating enhanced integrated information transmission. To investigate how disinhibition-mediated different inhibitory circuit modulates multisensory integration, we differentiated SST+ inhibitory circuits by intra-columnar feedback (intra-FBI), intra-columnar feedforward (intra-FFI), and cross-columnar feedforward inhibition (cross-FFI). When we fedin vivospike patterns into these models, we found that MIratewas highest in intra-FFI, whereas MI for spike timing was highest with intra-FBI, implying distinct roles in neural coding.

Discussion
Our results demonstrate that disinhibitory circuits facilitate multisensory integration by dynamical modulation of long-range cross-modal interactions between A1 and V1. Specifically, our findings reveal that the intra-FFI circuit was associated with firing rates whereas intra-FBI circuit enhanced information encoded in spike timings. This suggests that distinct disinhibitory circuits selectively integrate multisensory information through different neural coding strategies. Taken together, these findings indicate that distinct disinhibitory network motifs dynamically modulate multisensory integration and may serve as a key mechanism inin vivomultisensory processing.

Acknowledgements
This research was supported by the KIST Institutional Program (2E33561) and the National R&D Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Science and ICT (NRF-2021R1C1C2012843). J.-H. Han was supported by the MSIT, Korea, under the ITRC support program (IITP-2025-RS-2022-00156225) supervised by the IITP and by the NRF grant (No. RS-2024-00415812).
References
1. https://doi.org/10.1038/ncomms12815
2. https://doi.org/10.1016/j.conb.2018.01.002
3. https://doi.org/10.1016/j.tins.2021.04.009
4. https://doi.org/10.1007/s10827-017-0669-1

5. https://doi.org/10.1016/j.neuron.2016.01.027

Speakers

Seungyoun Lee

Hyun Jae Jang

Kyujin Kang

Yebeen Yoon

Jae-Ho Han

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P164: Event-Driven Financial Decision Making via Spiking Neural Networks: Neuromorphic-Inspired Approach

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P164 Event-Driven Financial Decision Making via Spiking Neural Networks: Neuromorphic-Inspired Approach

Tae-hoon Lee1, Hoon-hee Kim*2
1 Department of Data Engineering, Pukyong National University, Busan, South Korea
2 Department of Computer Engineering and Artificial Intelligence, Pukyong National University, Busan, South Korea
*Email: h2kim@pknu.ac.kr
Introduction

Spiking Neural Networks (SNNs) are well-suited for financial decision-making due to their ability to capture temporal dynamics and process information in an event-driven manner. In volatile markets, price movements can be sudden and irregular, making asynchronous event-based processing critical for timely responses. SNNs naturally handle such inputs, modeling temporal patterns more effectively than traditional neural networks. In this study, we integrate SNNs with a Genetic Algorithm (GA) for feature selection and parameter optimization, and a Support Vector Machine (SVM) for decision-making. This pipeline leverages the adaptive, event-driven processing of SNNs to improve stock market prediction and trading decisions.

Methods
For our experiments, we used historical data from the top 20 S&P 500 stocks, encompassing bull, bear, and volatile market conditions. Price data were transformed into multiple technical indicators (e.g., moving averages, RSI). A GA then optimized the indicator parameters and selected the most predictive features. Next, the time-series features were encoded into spike trains via rate coding with a fixed time window and fed into an SNN composed of Leaky Integrate-and-Fire neurons. The SNN processed the temporal patterns, and its spiking outputs were summarized (e.g., as spike counts over time). These features were then passed to an SVM for final classification of the trading action.

Results
In backtesting, the SNN-based framework surpassed the buy-and-hold strategy across multiple market regimes, demonstrating higher predictive accuracy and stronger trading returns. This performance gap was especially evident during volatile market phases, where passive buy-and-hold approaches often struggled to adapt. By capitalizing on the event-driven nature of spiking neurons, our system reacted swiftly to abrupt price swings, refining its signals in real time and thus helping to mitigate slippage and transaction costs. Overall, these findings highlight the neuromorphic framework’s resilience and effectiveness, suggesting it can outperform simpler investment strategies under diverse market conditions.

Discussion
This work demonstrates the potential of neuromorphic computing in financial decision-making. The SNN-based approach offers adaptive, event-driven processing suited to volatile markets, while its reservoir-like architecture (with only the output classifier trained) reduces computational complexity. In addition, the model exhibits robustness to noisy market data and regime shifts. However, limitations remain: the approach relies on a predefined rate-coding scheme, and the hybrid design combining a spiking network with an external classifier is not end-to-end. Future research can explore improved encoding methods and end-to-end spiking models, as well as deployment on neuromorphic hardware for faster, energy-efficient execution.

Figure 1. Flowchart: Stock data and optimized technical indicators are converted into spike trains for the spiking neural network (SNN), whose outputs feed into a classifier for trading decisions
Acknowledgements
This study was supported by the National Police Agency and the Ministry of Science, ICT & Future Planning (2024-SCPO-B-0130), the National Research Foundation of Korea grant funded by the Korea government (RS-2023-00242528), the National Program for Excellence in SW, supervised by the IITP(Institute of Information & communications Technology Planing & Evaluation) in 2025(2024-0-00018)
References
[1] Maass, W. (1997). Networks of Spiking Neurons: The Third Generation of Neural Network Models.Neural Networks, 10(9), 1659–1671. https://doi.org/10.1016/S0893-6080(97)00011-7
[2] Holland, J. H. (1992). Genetic algorithms.Scientific American, 267(1), 66–73.http://www.jstor.org/stable/24939139
[3] Lin, X., Yang, Z., & Song, Y. (2011). Intelligent stock trading system based on improved technical analysis and Echo State Network.Expert Systems with Applications, 38(9), 11347–11354. https://doi.org/10.1016/j.eswa.2011.03.001

Speakers

Hoon-hee Kim

Taehoon Lee

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P165: Incorporation of Neuromodulation into Predictive-Coding Spiking Neural Networks

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P165 Incorporation of Neuromodulation into Predictive-Coding Spiking Neural Networks

Yelim Lee1, Dongmyeong Lee1, Hae-Jeong Park*1,2,3,4

1Department of Nuclear Medicine, Graduate School of Medical Science, Brain Korea 21 Project, Yonsei University College of Medicine, Seoul, Republic of Korea
2 Department of Nuclear Medicine, Severance Hospital,Seoul, Republic of Korea
3Department of Cognitive Science, Yonsei University, Seoul, Republic of Korea
4Center for Systems and Translational Brain Sciences, Institute of Human Complexity and Systems Science, Yonsei University, Seoul, Republic of Korea

*Email: parkhj@yonsei.ac.kr

Introduction

Neuromodulation is often considered to enhance the selection mechanism in the brains of living organisms, prioritizing processing of inputs relevant to their goals. By modifying effective synaptic strength and altering firing properties, neuromodulators engage various cellular mechanisms, leading to a dynamic reconfiguration of neural circuits. Adjusting a target neuron’s excitability is one mechanism for enabling attentional effects. This research explores how this mechanism enhances predictive coding and learning in a spiking neural network (SNN) with two-compartment neurons, focusing on classification ability and internal representations in hidden layers with top-down signals.

Methods
The network includes one input layer, one output layer, and three fully connected hidden layers with feedback and feedforward connections. The dynamics of the hidden neurons are based on the Adaptive Leaky-Integrate-and-Fire (ALIF) model from Zhang and Bohte's previous work. The dendritic compartment of the hidden neurons integrate inputs from higher regions, and the somatic compartment integrates input from lower areas. To implement the neuromodulation effect on the hidden neurons, we introduced a new top-down attention connection from the higher layer to the lower layer. This adjustment enables modifying the target neuron’s excitability by dynamically altering the baseline firing threshold. We used a spiking MNIST image as input data, modifying the original MNIST dataset to provide spiking input over time. Additionally, we created multiple variations of the MNIST dataset, introducing noise or making occluded or overlapped images, to provide the ambiguous context.
Results
We performed image classification tasks with the MNIST dataset, achieving a high accuracy for the original set and highly noisy test data set. We analyzed the uncertainty of output neurons by tracking their membrane potential for each digit class, noting increased firing for the correct class despite initial uncertainty. To assess predictive coding, we evaluated each hidden layer's internal representation by decoding the spiking activity. This involved no inputs or half-occluded inputs while clamping the output neuron’s membrane potential to a specific class. The results showed successful digit representation in spiking activities, especially with applied modulation weights, compared to the previous model.
Discussion
Clarifying important information in uncertain contexts improves with appropriate attention and prediction. This study suggests that neuromodulation enhances hierarchical encoding and learning in SNN during ambiguous scenarios. The model maintained high classification accuracy even in noisy and occluded conditions, and the internal representation, along with reduced uncertainty of output neurons, aligns with predictive coding principles, where top-down modulation refines internal representations.

Acknowledgements
This work was supported by the National Research Foundation of Korea(NRF) grant funded by the Korea government(MSIT) (NO. 2023R1A2C200621711)
References
1.Desimone, R., & Duncan, J. (1995). Neural mechanisms of selective visual attention. Annual review of neuroscience, 18(1), 193-222.
2.Marder, E. (2012). Neuromodulation of neuronal circuits: back to the future. Neuron, 76(1), 1-11.
3.Thiele, A., & Bellgrove, M. A. (2018). Neuromodulation of attention. Neuron, 97(4), 769-785.
4.Zhang, M., & Bohte, S. M. (2024). Energy Optimization Induces Predictive-coding Properties in a Multicompartment Spiking Neural Network Model. bioRxiv, 2024-01.

Speakers

Hae-Jeong Park

DongMyeong Lee

Yelim Lee

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P166: Leveraging neural modeling of channelopathies to elucidate neural mechanisms underlying neurodevelopmental disorders

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P166 Leveraging neural modeling of channelopathies to elucidate neural mechanisms underlying neurodevelopmental disorders

Molly Leitner*1, Roman Baravalle1,James Chen1,Timothy Fenton3, Roy Ben-Shalom3, Salvador Dura-Bernal1,2

1Department of Physiology and Pharmacology, SUNY Downstate Health Sciences University, Brooklyn, NY, USA
2Center for Biomedical Imaging & Neuromodulation, Nathan Kline Institute for Psychiatric Research, Orangeburg, NY, USA
3Neurology Department, University of California Davis, Davis, CA, USA

*Email: molly.leitner@downstate.edu
Introduction

Neurodevelopmental disorders (NDDs), such as epilepsy, autism spectrum disorder, and developmental delays, present with considerable clinical variability and often impair social interactions, speech, and cognitive development. A key feature of these disorders is an imbalance in excitatory/inhibitory (E/I) input, which disrupts neuronal circuit function during development. Brain channelopathies, where neuronal ion channel activity is altered, provide an ideal model for studying E/I imbalance, as their effects can be directly linked to neuronal excitability. Ion channels are crucial in generating electrical activity in neurons, and disruptions to this activity are strongly associated with NDDs [1].

Methods
Studying channelopathies at the single-cell level is well-established, however, investigating the impact of specific channel mutations on neuronal circuits requires more complex approaches.By utilizing a previously developed primary motor cortex (M1) model built using NetPyNE and NEURON, we employ large-scale, highly detailed biophysical neuronal simulations to examine how channel mutations influence individual and network neuronal activity [2].

Results
These simulations offer a mechanistic understanding of how channelopathies contribute to E/I imbalance and the pathology of NDDs. Through the M1 cortical column simulation, we measure the effects of biophysical changes in ion channels on network excitability and neuronal firing patterns, providing insights into the pathophysiology of simulated channelopathies.
Discussion
This model not only serves as a tool for investigating specific channelopathy cases but also enables the exploration of pharmacological agents aimed at restoring E/I balance. Ultimately, this approach will enhance our understanding of targeted therapeutic strategies for alleviating disease symptoms and may uncover novel treatments with clinical potential.

Acknowledgements
This work was supported by the Hartwell Foundation through an Individual Biomedical Research Award. The authors gratefully acknowledge the foundation’s commitment to innovative pediatric research and its generous support of our project.
References
● Spratt PWE, Ben-Shalom R, Keeshen CM, Burke KJ Jr, Clarkson RL, Sanders SJ, Bender KJ. The Autism-Associated Gene Scn2a Contributes to Dendritic Excitability and Synaptic Function in the Prefrontal Cortex. Neuron. 2019 Aug 21;103(4):673-685.e5. doi: 10.1016/j.neuron.2019.05.037.
● Dura-Bernal S, Neymotin SA, Suter BA, Dacre J, Moreira JVS, Urdapilleta E, Schiemann J, Duguid I, Shepherd GMG, Lytton WW. Multiscale model of primary motor cortex circuits predicts in vivo cell-type-specific, behavioral state-dependent dynamics. Cell Rep. 2023 Jun 27;42(6):112574. doi: 10.1016/j.celrep.2023.112574.

Speakers

Roman Baravalle

Salvador Dura-Bernal

Molly Leitner

James Chen

Timothy Fenton

Roy Ben-Shalom

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P167: Active NMDARs expand input rate sensitivity into high-conductance states

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P167 Active NMDARs expand input rate sensitivity into high-conductance states

Movitz Lenninger*1, Pawel Herman1, Mikael Skoglund1, Arvind Kumar1

1 School of Electrical Engineering and Computer Science, KTH Royal Institute of Technology, Stockholm, Sweden
*Email: movitzle@kth.se

Introduction

A single cell has thousands of synapses distributed across the surface, predominantly along the dendritic tree [1]. Thus, in an active state, thousands of inputs can target a single cell leading to what is known as the high conductance state [2]. During such states, both input resistance and the effective membrane time constant are markedly reduced [3]. Paradoxically, high-conductance states can also lead to a reduction of postsynaptic activity [4,5]. Here, we show, using single-cell simulations of thick-tuft layer 5 (TTL5) pyramidal cells, that the voltage dependence of NMDA receptors (NMDARs), a ubiquitous feature in the brain, can increase excitability in high-conductance states – providing sensitivity to a larger range of inputs.

Methods
We simulated a previously published reconstructed morphology of a rat TTL5 pyramidal cell [6]. We randomly distribute 5000 excitatory and 2500 inhibitory synapses uniformly according to the membrane surface areas of the dendritic segments (Figure 1a). Inputs are sampled from independent Poisson processes. In all cases, we optimize the inhibitory input rate to keep the somatic potential fluctuating around -60 mV. To study the role of active NMDARs, we consider three scenarios: synapses contain (1) only AMPA receptors (AMPARs), (2) both AMPARs and active NMDARs, and (3) both AMPARs and passive NMDARs. Unless otherwise stated, we use an NMDA-AMPA ratio of 1.6. In all cases, the integrated conductance per input is normalized to ~5.9 nS∙ms.

Results
First, we compare the input resistances across three input conditions. The input resistance decreases with increasing inputs for all three synaptic types but is consistently larger with active NMDARs (Figure 1b). Second, we compare the output firing rates (FRs) across a large range of inputs. For low and intermediate inputs, the output FRs are similar across all synaptic types (Figure 1c). However, for high inputs, output gain is only maintained with active NMDARs. Furthermore, the coefficient of variation of the interspike intervals is typically higher for active NMDARs, indicating more irregular firing (Figure 1d). Third, varying the NMDA-AMPA ratio reveals that this is a graded property of active NMDARs (Figure 1e-f).

Discussion
A key property of dendrites is to integrate pre-synaptic inputs. Active conductance can significantly alter the summation compared to passive dendrites [1]. Previous studies have, for example, linked active NMDARs to increased sequence discrimination [7] and increased coupling between tuft and soma [8]. Our work suggests active NMDARs might also be crucial for maintaining large postsynaptic activity under high input conditions, expanding the range of input sensitivity. Our work does not exclude the possibility of intrinsic voltage-gated ion channels further contributing to increased excitability under presynaptic activity [9]. It remains to study the possible interaction of such intrinsicconductanceswith active NMDRs.

Figure 1. a) Morphology of cell with 500 randomly distributed synapses. b) Estimated input resistances during three input conditions. c) Input-output transfer function of firing rates (lines). Shaded areas show the standard deviation (across bins of 1 second). d) CVs of the ISIs. Color codes in panels c-d) same as in b). e-f) Output firing rates and CVs for a range of NMDA-AMPA ratios with active NMDARs.
AcknowledgementsN/A
References
[1]https://doi.org/10.1146/annurev.neuro.28.061604.135703
[2]https://doi.org/10.1038/nrn1198
[3]https://doi.org/10.1073/pnas.88.24.11569
[4]https://doi.org/10.1523/JNEUROSCI.3349-03.2004
[5]https://doi.org/10.1103/PhysRevX.12.011044
[6]https://doi.org/10.1371/journal.pcbi.1002107
[7]https://doi.org/10.1126/science.1189664
[8]https://doi.org/10.1038/nn.3646
[9]https://doi.org/10.1016/J.NEURON.2020.04.001

Speakers

Pawel Herman

Movitz Lenninger

Arvind Kumar

Mikael Skoglund

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P168: Evaluating Effective Connectivity and Control Theory to Understand rTMS-Induced Network Effects

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P168 Evaluating Effective Connectivity and Control Theory to Understand rTMS-Induced Network Effects

Riccardo Leone*1,2,3, Michele Allegra4, Xenia Kobeleva1,2
1 Computational Neurology Group, Ruhr University Bochum, 44801, Bochum, Germany.
2 Faculty of Medicine, University of Bonn, 53127, Bonn, Germany.
3 German Center for Neurodegenerative Diseases (DZNE), 53127, Bonn, Germany.
4 Padova Neuroscience Center, University of Padova,35129,Padova, Italy

* Email: riccardoleone1991@gmail.com
Introduction

Computational neuroscience might contribute to a better understanding of neurostimulation by modeling its effects on brain networks. Effective connectivity (EC) and EC-based network control theory could provide a theory-driven framework for elucidating neurostimulation-induced network effects [1]. We thus tested whether EC and control energy could explain changes in resting-state fMRI (rs-fMRI) metrics induced by repetitive transcranial magnetic stimulation (rTMS). We hypothesized that EC and control energy would outperform functional connectivity (FC) and structural connectivity (SC) in explaining rTMS effects.

Methods
Twenty-one subjects received inhibitory 1Hz rTMS (20 min) at frontal, occipital or temporo-parietal sites, with rs-fMRI acquired pre- and post-stimulation. Whole-brain EC was estimated using regression Dynamic Causal Modeling. Control energy from the stimulated node (i.e., driver node) to each downstream target node was computed from the EC model. We quantified rTMS effects at a node level as pre- vs post- changes in: i) FC with the driver region, ii) amplitude of low-frequency fluctuations (ALFF), and iii) nodal FC strength with the whole brain. We correlated these changes with a series of pre-stimulus predictors: SC, FC, EC between each target and the driver node, and energy needed to control each target from the driver node.

Results
rTMS generally reduced whole-brain FC with each stimulated driver node, as well as ALFF, and nodal FC strength, with frontal stimulation yielding more widespread effects. EC and control energy showed significant correlations with the change in FC with the driver node and nodal FC strength. Nonetheless, significant associations of similar or greater magnitude were also observed with simple FC, thus failing to demonstrate a clear advantage of EC and EC-based control energy to evaluate rTMS-induced effects. Changes in ALFF were not significantly correlated with any pre-TMS variable.
Discussion
Contrary to our main hypothesis, EC and EC-based control energy did not provide significantly better explanations of 1Hz rTMS-induced changes compared to model-agnostic FC. Our results question the current utility of EC and EC-based control theory models for understanding the effects of 1-Hz rTMS on brain networks. Given the complex interplay of neurobiological processes induced by rTMS that are not directly linked to the network spread of TMS pulses (e.g., synaptic plasticity), future work should implement EC and EC-based control energy to explain the effects of simpler protocols of neurostimulation.

Acknowledgements

References
1. Manjunatha KKH, Baron G, Benozzo D, Silvestri E, Corbetta M, Chiuso A, et al. (2024) Controlling target brain regions by optimal selection of input nodes.PLoS Comput Biol20(1): e1011274. https://doi.org/10.1371/journal.pcbi.1011274

Speakers

Michele Allegra

Riccardo Leone

Xenia Kobeleva

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P169: Anesthesia modulates the system-wide contributions of identified head neurons in C. elegans

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P169 Anesthesia modulates the system-wide contributions of identified head neurons in C. elegans

Avraham Lepsky*¹, Andrew Chang², Chris Connor³, Chris Gabel³

¹ Graduate Program for Neuroscience, Boston University, Boston, United States
² Graduate Program in Physiology, Boston University, Boston, United States
³ Department of Biophysics, Boston University, Boston, United States

*Email: avil@bu.edu
Introduction

While anesthesia has similar effects in the brains of animals ranging from the nematode C. elegans to humans, the mechanism by which various anesthetic agents work remains largely unknown. C. elegans has been identified as a tractable model for studying anesthesia due to progressing behavioral deficits from increased anesthesia concentration, genetic susceptibility analogous to mammals, and an annotated neuroconnectome [1]. Isoflurane is a volatile anesthetic that induces general anesthesia; previous work has found that isoflurane anesthesia in C. elegans caused marked dyssynchrony of neuron dynamics (as measured by a decrease in the cumulative variance explained by the top 3 principal components in neuronal activity) [2].

Methods
We employed C. elegans worms expressing the NeuroPAL transgene, providing a fluorescent color map for identification of neurons within the known connectome of the C. elegans nervous system [3]. Using light sheet microscopy performed by a dual inverted selective plane illumination microscope (DISPIM), we measured activity of 120 individual head neurons of the NeuroPAL worms via fluorescence imaging of the calcium sensitive GCaMP reporter. We imaged for 20 minutes at 2Hz. We performed principal component analysis (PCA) on the measured 120 neuron activity dynamics, following previous attempts at ascribing a neural manifold to C. elegans behavior [4] with the added information of neuron identification.
Results
Analysis of neuronal activity across worms at various isoflurane levels identified 10 neurons with a statistically significant change in PCA magnitude between 0 and 2% isoflurane and 17 neurons between 0 and 4%. No obvious receptor or functional identity marker was shared by all statistically significant neurons.
Discussion
We identified a list of neurons whose contributions to the system’s activity are most significantly modulated by changing isoflurane concentration. Because the connectome of C. elegans has been established, the anatomical properties of the neurons can be compared to their functional properties to establish a mechanistic understanding of the systemic changes induced by isoflurane. Connectomic spiking neuron models and other biophysical models can then be used to make predictions linking the molecular and behavioral properties of anesthetic agents.

Acknowledgements
Thank you to the Graduate Program of Neuroscience, under the direction of Dr. Shelly J. Russek and Sandi Grasso, for providing such a nurturing community.
Funding was generously awarded through a T32 grant.
References
●
Rajaram, S., … & Morgan, P. G. (1999). A stomatin and a degenerin interact to control anesthetic sensitivity in Caenorhabditis elegans. Genetics, 153(4), 1673–1682.
●
Awal, M. R., … & Connor, C. W. (2020). The collapse of global neuronal states in C. elegans under isoflurane anesthesia. Anesthesiology, 133(1), 133.
●
Yemini, E., ... & Hobert, O. (2021). NeuroPAL: a multicolor atlas for whole-brain neuronal identification in C. elegans. Cell, 184(1), 272-288.
●
Kato, S., ... & Zimmer, M. (2015). Global brain dynamics embed the motor command sequence of Caenorhabditis elegans. Cell, 163(3), 656-669.

Speakers

Avraham Lepsky

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P170: Competition between symmetric and antisymmetric connections in cortical networks

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P170 Competition between symmetric and antisymmetric connections in cortical networks

Dong Li*1,Claus C. Hilgetag1,2

1Institut für Computational Neuroscience, Universitätsklinikum Hamburg-Eppendorf (UKE), 20246 Hamburg, Germany

2Department of Health Sciences, Boston University, 02215 Boston, USA

*Email: d.li@uke.de

Introduction

The pairwise correlation of neural activity directly and significantly influences neural network performance across various cognitive tasks [1, 2]. While tasks such as working memory require low correlation levels [3], others, like motor actions, rely on higher correlation levels [1]. These correlation patterns are highly sensitive to network structure and neural plasticity [4-6]. However, understanding how neural networks dynamically balance tasks with differing correlation demands, and how distinct brain networks are structurally optimized for specific functions remains a major challenge.

Methods
We simulate linear and spiking models to investigate the impact of symmetric and antisymmetric connections on neural network dynamics. The linear model, equipped with a control parameter that adjusts the relative intensity of these connections, captures fundamental mechanisms, which shape pairwise correlations and influence network performance across cognitive tasks. To quantify the competition between symmetric and antisymmetric connections, we introduce two indices from global and local perspectives. Using these indices, we further examine how synaptic plasticity modulates the relative intensity of these connections. Finally, we employ the spiking model to explore how bio-plausible neural networks implement this competition.
Results
Antisymmetric connections naturally reduce pairwise correlations, facilitating cognitive tasks that require maximal information processing, such as working memory. In contrast, symmetric connections enhance pairwise correlations, supporting other functions, such as enabling the network to generate reliable responses to external inputs. The competition between antisymmetric and symmetric connections can be easily modulated by spike-timing-dependent plasticity (STDP) with antisymmetric and symmetric kernels, respectively. In bio-plausible networks, this competition is particularly shaped by the structured, non-random organization of excitatory and inhibitory connections.
Discussion
Every connection matrix can be decomposed into symmetric and antisymmetric components with varying relative intensities. This work reveals how the competition between these components modulates neural correlations and facilitates distinct functions. Temporally, this competition is dynamically regulated by synaptic plasticity. Spatially, in comparison to indirect experimental evidence, our analysis actually also allows the discussion of the layer-specific distribution of these relative intensities. These findings provide a new perspective on how brain functions are segregated across both time and space.

Acknowledgements
This work was in part founded by DFGTRR-169 (A2) andSFB 936 (A1/Z3).
References
[1]https://doi.org/10.1038/nrn1888
[2]Von Der Malsburg, C. (1994). The correlation theory of brain function. In Models of neural networks: Temporal aspects of coding and information processing in biological systems (pp. 95-119). New York, NY: Springer New York.
[3]https://doi.org/10.1016/j.dcn.2025.101541
[4]https://doi.org/10.1016/0893-6080(94)00108-X
[5]https://doi.org/10.1126/science.1211095
[6]https://doi.org/10.1162/NECO_a_00451

Speakers

Claus Hilgetag

Dong Li

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P171: Partial Information Decomposition of amplitude and phase stimulus encoding in oscillator models

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P171 Partial Information Decomposition of amplitude and phase stimulus encoding in oscillator models

V. LIMA¹*, D. MARINAZZO², A. BROVELLI¹,

1. Institut de neurosciences de la Timone, Aix Marseille Université, UMR 7289 CNRS, 13005, Marseille, France.
2. Faculty of Psychology and Educational Sciences, Department of Data Analysis, University of Ghent, Ghent, Belgium

* vinicius.lima.cordeiro@gmail.com

Introduction

Synchrony between oscillatory activity is thought to be the primary mechanism enabling widespread cortical regions to route information [1]. Such a mechanism would require either oscillations between a target and a sending area to increase their amplitude while maintaining a stable phase relationship or to shift their phase difference when a stimulus is presented [2]. Nonetheless, whether the “communication” established between the pair of areas can be used to encode stimulus-specific information remains unclear.

Methods
To address this question, we construct a whole-brain model in which nodes are connected using macaque structural connectivity [3], and their dynamics are governed by the Stuart-Landau (SL) model [4]. The SL model describes nonlinear oscillators near a Hopf bifurcation and models both the evolution of their amplitude and phase terms. In addition to enabling the characterization of interactions in terms of phase and/or amplitude, the distance to the Hopf bifurcation is controlled by a single parameter,a, which determines the stability of the oscillations:a< 0 leads to transient oscillations, whereasa≥ 0 results in stable oscillations, allowing to explore the role of both types of activity in stimulus encoding [5].To disentangle phase and amplitude encoding in the model, we use the framework of partial information decomposition (PID) [6] to estimate the information that the phase and amplitude components of simulated neuronal activity uniquely carry about the stimulus. Briefly, for two nodes indexed byjandk, we consider the product of their amplitude termsAjk, phase differencejk, and stimulusS. The three-variable PID allows us to decompose their total mutual informationI(S;Ajk,jk)into terms representing how they encode the stimulus redundantly or synergistically, as well as the unique information contained in the amplitude and phase interactions. Additionally, this framework could be extended to study non-dyadic interactions by operating at edge rather than node level [7]. In this case, PID decomposition is performed between two edge time series, each given byEjk=Ajkejk, allowing to either decompose the following mutual information terms:I(S;Ajk,Aml),I(S;Ajk,ml), andI(S;jk,ml)or performing multivariate PID.

Results

In the whole-brain model, we found that even though the stimulus is generally encoded by the signals’ amplitude, in areas that are hierarchically far apart, the initial encoding in amplitude is later found in the phase relation between the two areas in a weaker but more persistent form. These effects highly depend on the nodes’ dynamics and are most favorable when they exhibit transient oscillations (a < 0).

Discussion
Introducing a scaling of the natural oscillation frequency also appeared to enhance the effect, suggesting that different time scales across the cortex may promote the establishment of functional coupling through phase synchrony [8].

Acknowledgements
None
References
●
doi.org/10.1016/j.neuron.2015.09.034

●
10.1016/j.neuron.2023.03.015

●
10.1093/cercor/bhs270

●
10.1038/s41598-024-53105-0

●
10.1016/j.tics.2024.09.013Get rights and content

●
arxiv.org/abs/1004.2515

●
10.1038/s41593-020-00719-y

●
10.1073/pnas.1402773111

Speakers

Andrea Brovelli

Vinicius Lima Cordeiro

Daniele Marinazzo

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P172: Scalable Computational Modeling of Neuron-Astrocyte Interactions in NEST

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P172 Scalable Computational Modeling of Neuron-Astrocyte Interactions in NEST

Marja-Leena Linne*1, Han-Jia Jiang2,3, Jugoslava Aćimović1, Tiina Manninen1, Iiro Ahokainen1, Jonas Stapmanns2,4, Mikko Lehtimäki1, Markus Diesmann2,4,5, Sacha J. van Albada2,3, Hans Ekkehard Plesser2,6,7
1Faculty of Medicine and Health Technology, Tampere University, Tampere, Finland
2Institute for Advanced Simulation (IAS-6), Jülich Research Centre, Jülich, Germany
3Institute of Zoology, Faculty of Mathematics and Natural Sciences, University of Cologne, Cologne, Germany
4Department of Physics, Faculty 1, RWTH Aachen University, Aachen, Germany
5Department of Psychiatry, Psychotherapy and Psychosomatics, School of Medicine, RWTH Aachen University, Aachen, Germany
6Department of Data Science, Faculty of Science and Technology, Norwegian University of Life Sciences, Ås, Norway
7Käte Hamburger Kolleg: Cultures of Research (c:o/re), RWTH Aachen University, Aachen, Germany

*Email: marja-leena.linne@tuni.fi
Introduction

Astrocytes play a key role in modulating synaptic activity and network dynamics, yet large-scale models incorporating neuron-astrocyte interactions remain scarce [1]. This study introduces a novel NEST-based [2] simulation framework to model tripartite connectivity, where astrocytes interact with both presynaptic and postsynaptic neurons, extending traditional binary synaptic architectures. By integrating astrocytic calcium signaling and astrocyte-induced synaptic currents (SICs), the model enables dynamic modulation of neuronal activity, offering insights into the role of astrocytes in neural computation.
Methods
Our implementation integrates astrocytic calcium dynamics and SICs within a scalable, parameterized framework. The model allows controlled modulation of astrocytic influence, capturing transitions between asynchronous and synchronized neuronal states. Simulation scalability was assessed through strong and weak scaling benchmarks, leveraging parallel computing for network performance evaluation. Strong scaling benchmarks tested performance under fixed model size while increasing computing resources. Weak scaling benchmarks examined proportional upscaling of model size and computational power. These benchmarks evaluated network connection times, state propagation efficiency, and computational cost across different neuron-astrocyte configurations.
Results
Benchmark results show efficient parallel execution of the reference implementation [3]. Strong scaling benchmarks show that increasing computing resources reduces network connection and state propagation times. Weak scaling benchmarks reveal a moderate increase in communication time for processes like spike delivery and SIC delivery, yet overall performance remains robust against changes in model size and connectivity scheme. In this study, we validate the framework’s scalability to at least 1 million cells through benchmarking experiments, leveraging distributed computing for efficient simulation of large-scale neuron-glia networks.
Discussion
By providing a computationally accessible and reproducible tool for studying neuron-astrocyte interactions, this framework sets the stage for investigating glial contributions to synaptic modulation, network coordination, and their roles in neurological disorders. The integration of tripartite connectivity into NEST offers a versatile platform for modeling astrocytic regulation of neural circuits, advancing both fundamental neuroscience and applied computational modeling.

Acknowledgements
EU Horizon 2020 No. 945539 (Human Brain Project SGA3) to SJvA and M-LL. SGA3 Partnering Project (AstroNeuronNets) to JA and SJvA. EU Horizon Europe No. 101147319 (EBRAINS 2.0 Project) to SJvA and M-LL. HiRSE PS to SJvA. Research Council of Finland, Nos. 326494, 326495, 345280, and 355256, to TM, and 297893 and 318879 to M-LL. BMBF No. 01UK2104 (KHK c:o/re) to HEP.
References
[1]Manninen, T., Aćimović, J., Linne, M.-L. (2023). Analysis of Network Models with Neuron-Astrocyte Interactions. Neuroinformatics, 21(2), 375-406.https://doi.org/10.1007/s12021-023-09622-w
[2]Graber, S., Mitchell, J., Kurth, A.C., Terhorst, D., Skaar, J.E.W., Schöfmann, C.M., et al.(2024). NEST 3.8.https://zenodo.org/records/12624784

[3]Jiang, H.-J.,Aćimović, J., Manninen, T., Ahokainen, I., Stapmanns, J., Lehtimäki, M., et al.(2024). Modeling neuron-astrocyte interactions in neural networks using distributed simulation. bioRxiv.https://doi.org/10.1101/2024.11.11.622953

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P173: Large-scale Neural Network Model of the Human Cerebral Cortex Based on High-spatial-resolution Imaging Data

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P173 Large-scale Neural Network Model of the Human Cerebral Cortex Based on High-spatial-resolution Imaging Data

Chang Liu1, Dahui Wang*1, Yuxiu Shao*1

1School of Systems Science, Beijing Normal University, Beijing, China
*Email: wangdh@bnu.edu.cn (DW); shaoyx@bnu.edu.cn (YS)
Introduction

Large-scale brain models have received interest for their ability to probe complex dynamical phenomena. However, large-scale models guided by high-spatial-resolution imaging data remain largely underexplored. We develop a comprehensive large-scale model of the human cerebral cortex, utilizing the recently released data on receptor density[1] and white-matter connectivity[2]. Furthermore, we refine undirected white-matter connectivity into directed connectivity using tracer data[3]. Our model replicates the characteristic spatio-temporal patterns of whole-brain activities observed experimentally during the resting state, enabling a deeper exploration of the interplay between anatomical structure, dynamics, and potential functional roles.
Methods
Our network comprises about 60k vertices, each modeled as a microcircuit of coupled excitatory and inhibitory populations connected via AMPA, NMDA, and GABA synapses, exhibiting Wilson-Cowan type dynamics[4]. Intra-vertex connection strengths are proportional to receptor density[1]. Inter-vertex connections are derived from anatomical fiber data, obtained via dMRI tractography at vertex resolution[2], and averaged across 255 unrelated healthy individuals (Fig. 1A). Since this anatomical data is undirected, we redistribute fiber bundles between vertices using directed macaque neocortex tracer data[3].
Results
The simulation results demonstrate that the averaged firing rate (FR) across all vertices is around 3Hz[5]. Interconnected vertices show reduced correlation between FR and the excitatory-inhibitory receptor density ratio compared to independent vertices (Fig. 1B). Beta-band peak frequency exhibits a posterior-anterior gradient, which is disrupted by shuffling the spatial distribution of the AMPA-NMDA receptor ratio (Fig. 1C). The projection of power spectral density and FR onto the first principal component positively correlate with T1w/T2w (Fig. 1D, 1E). These findings align with experimental observations[6–8]. Moreover, asymmetric connectivity induces traveling waves, with sinks exhibiting higher FR than surrounding vertices (Fig. 1F).
Discussion
Our large-scale brain model with high-spatial-resolution not only introduces a novel approach to understanding the computational mechanisms of the brain but also offers critical insights into the neural dynamic mechanisms underlying cognitive dysfunction and mental disorders. However, our model still has some limitations: we directly assume synaptic strength is proportional to receptor density; estimate the directed-weighted connections based on coarse matching of macaque-human brain areas; and omit signal propagation delays. Future work will focus on simulating the information transmission across the cortex, exploring how this model can enhance our understanding of brain function and support the development of therapeutic strategies.

Figure 1. Fig 1: (A) Schematic. (B) Relationship between mean FR and E:I ratio. Blue: independent vertices. Red: interconnected vertices. (C) Dependency between the vertex’s location along the posterior-anterior axis. Blue: original. Pink: shuffled density ratio. (D-E) Correlation of model PSD PC1 maps (D) with T1w/T2w, and model FR PC1 maps (E). (F) Sinks displaying higher FR than surrounding vertices.
Acknowledgements
This work was supported by NSFC (No.32171094 to D.W., No.32400936 to Y.S.) and National Key R&D Program of China (2019YFA0709503 to D.W.) and International Brain Research Organization Early Career Award (to Y.S.).
References
1. https://doi.org/10.1038/s41593-022-01186-3
2. https://doi.org/10.1016/j.neuroimage.2020.117695
3. https://doi.org/10.1093/cercor/bhs270
4. https://doi.org/10.1523/JNEUROSCI.3733-05.2006
5. https://doi.org/10.1023/A:1011204814320
6. https://doi.org/10.7554/eLife.53715
7. https://doi.org/10.1016/j.neuron.2019.01.017
8. https://doi.org/10.1073/pnas.1608282113

Speakers

Chang Liu

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P174: Functional brain regions analysis using single-neuron morphology-driven reservoir network

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P174 Functional brain regions analysis using single-neuron morphology-driven reservoir network

Yuze Liu, Linus Manubens-Gil*, Hanchuan Peng*

Institute for Brain and Intelligence, Southeast University, Nanjing, China

* Email: linus.ma.gi@gmail.com

* Email: h@braintell.org
Introduction

The brain operates through network topology across brain regions and morphological diversity of neurons. Reservoir computing (RC), with its recurrent nonlinearly mapping, can accomplish temporal tasks [1], enabling network functional analysis. Previous work constructed reservoir using diffusion magnetic resonance imaging (dMRI)-derived connectivity matrices and proved randomness in weight signs improves network’s memory capacity (MC) [2]. However, limitations persist due to the macroscopic scale of connectome, leaving microscale neuronal contributions underexplored. Thus, we established a reservoir using mouse brain’s single-neuron full morphology tracings, analyzing its validity in exploring the variance of functional regions by MC task.

Methods
We used structural connectivity (SC) from [3]. The connectome data is 1,774 fully reconstructed mouse neurons registered to Allen Mouse Brain Common Coordinate Framework (CCFv3) [4]. We used hyperbolic tangent as nonlinear mapping. Input signal is sampled randomly, target signal is the delayed input. We fitted output to target signal via ridge regression and quantified performance by squared Pearson correlation coefficient. We constructed small-world networks with connection density approximating SC. We selected functional brain regions, e.g., LGd (Dorsal part of the lateral geniculate complex) and visual cortex regions as input/output nodes. We adjusted spectral radius to optimize connectivity weights for enhanced memory retention.
Results
We found that 1) based on uniform random connectivity weights, biologically wired networks with input-output nodes defined by functional regions slightly outperformed the Watts-Strogatz small-world network with random input-output node in MC task, confirming that single-neuron-derived network topology is relevant for the establishment of memories in RC; 2) we observed statistically significant differences in MC task performance for different thalamocortical integration of sensory modalities across diverse spectral radii ρ (76% of tested ρ values, 19/25, 0.1 ≤ ρ ≤ 5.0, Δρ = 0.2); independent t-test and Mann-Whitney U test, p<0.05;), suggesting morphological specificity of neuronal connections may underlie functional specialization.
Discussion
This study establishes a microscale framework linking single-neuron connectome to network functionality. Future work integrating generative models for scaling up the network, spiking neuronal dynamics, and modality-specific tasks could further dissect latent determinants of regional brain function.

Acknowledgements
This work was supported by the National Natural Science Foundation of China (NSFC) under Grant No. 32350410413 awarded to LMG.
References
1. https://doi.org/10.3389/fams.2024.1221051
2. https://doi.org/10.1109/IJCNN60899.2024.10650803
3. https://doi.org/10.1016/j.celrep.2024.113871
4. https://doi.org/10.1038/s41586-021-03941-1

Speakers

Linus Manubens-Gil

Hanchuan Peng

Yuze Liu

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P175: A Neurorobotic Framework for Exploring Locomotor Control Following Recovery from Thoracic Spinal Cord Injury

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P175 A Neurorobotic Framework for Exploring Locomotor Control Following Recovery from Thoracic Spinal Cord Injury

Andrew B. Lockhart*1, Huangrui Chu1, Shravan Tata Ramalingasetty1, Natalia A. Shevtsova1, David S.K. Magnuson2, Simon M. Danner1

1Department of Neurobiology and Anatomy, College of Medicine, Drexel University, Philadelphia, PA, USA

2Department of Neurological Surgery, University of Louisville, Louisville, KY, USA

*Email: abl73@drexel.edu

Introduction

Thoracic spinal cord contusion disrupts communication between the cervical and lumbar circuitry. Despite this, rats recover locomotor function, though at a reduced speed and with altered speed-dependent gait expression. Our previous computational model of spinal locomotor circuitry [2,3] reproduced the observed gait changes by linking them to impaired long propriospinal connectivity and lumbar circuitry reorganization, likely involving enhanced reliance on afferent feedback. To investigate the role of sensory feedback in locomotion and explore post-contusion reorganization, a neurorobotic model of quadrupedal locomotion was used in which the spinal circuitry was embedded in a body that interacted with the environment (Fig. 1).

Methods
We have expanded our previous neural network model of spinal locomotor circuitry to drive a simulated Unitree Go1 quadrupedal robot. The model includes four rhythm generators, one per limb, interconnected by commissural and long propriospinal neurons. Activity from each rhythm generator controls a pattern formation network that coordinates muscle activation in each limb. Hill-type muscles convert this activation into torque to actuate the motors and allow for calculation of proprioceptive feedback, which interacts with all levels of the spinal circuitry. Connection weights of proprioceptive, vestibular, and pattern forming neurons were optimized using covariance matrix adaptation evolution strategy to produce adaptive locomotion.
Results
The optimized model produces stable locomotion across a range of target speeds. Integration of muscle states and environmental information through proprioceptive, cutaneous, and vestibular neurons allows the model to traverse rough terrain consisting of variable slopes and ground friction. Preliminary simulation of thoracic contusion by reducing connection weights of inter-enlargement long propriospinal neurons results in altered gaits.

Discussion
The model provides a testbed for linking neuronal manipulations to changes in locomotion and behavior. By comparing locomotor gaits across models—those undergoing a second round of optimization post-contusion, those that have not, and experimental results from rats—we can identify and analyze critical neuronal connections involved in recovery. Using this approach, we will further investigate how circuit reorganization can contribute to locomotor recovery after thoracic spinal cord contusion.

Figure 1. Fig 1. A) The central locomotor circuit model for four limbs includes long propriospinal neurons connecting cervical and lumbar circuits adapted from Frigon 2017 [3]. B) Two-level rhythm and pattern formation circuitry for one limb. Motoneuron (MN) activity activates muscles (C) which actuate torque-controlled motors (D). Kinematics and kinetics are transformed into afferent feedback signals.
Acknowledgements
This work was supported by the National Institutes of Health (NIH) grantsR01NS112304, R01NS115900, andT32NS121768.
References
[1] Danner, S. M., et al. (2017). Computational modeling of spinal circuits controlling limb coordination and gaits in quadrupeds.eLife,6, e31050. https://doi.org/10.7554/eLife.31050
[2] Zhang, H., et al. (2022). The role of V3 neurons in speed-dependent interlimb coordination during locomotion in mice.eLife,11, e73424. https://doi.org/10.7554/eLife.73424
[3] Frigon, A. (2017). The neural control of interlimb coordination during mammalian locomotion.Journal of Neurophysiology,117(6), 2224–2241. https://doi.org/10.1152/jn.00978.2016

Speakers

Andrew Lockhart

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P176: Plastic Arbor: a modern simulation framework for synaptic plasticity – from single synapses to networks of morphological neurons

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P176 Plastic Arbor: a modern simulation framework for synaptic plasticity – from single synapses to networks of morphological neurons

Jannik Luboeinski*1,2,3, Sebastian Schmitt1,2, Shirin Shafiee1,2, Thorsten Hater4, Fabian Bösch5, Christian Tetzlaff1,2,3

1III. Institute of Physics – Biophysics, University of Göttingen, Germany
2Department for Neuro- and Sensory Physiology, University Medical Center Göttingen, Germany
3Campus Institute Data Science (CIDAS), Göttingen, Germany
4Jülich Supercomputing Centre, Forschungszentrum Jülich, Germany
5Swiss National Supercomputing Centre, ETH Zürich, Switzerland

*Email: jannik.luboeinski@med.uni-goettingen.de

Introduction
Arbor is a software library designed for the efficient simulation of large-scale networks of biological neurons with detailed morphological structures. It combines customizable neuronal and synaptic mechanisms with high-performance computing, enabling to use diverse backend architectures such as multi-core CPU and GPU systems [1] (also see Fig. 1a).
Synaptic plasticity processes play a vital role in cognitive functions, including learning and memory [2,3]. Recent studies have shown that intracellular molecular processes in dendrites significantly influence single-neuron dynamics [4,5]. However, for understanding how the complex interplay between dendrites and synaptic processes influences network dynamics, computational modeling is required.
Methods
To enable the modeling of large-scale networks of morphologically detailed neurons with diverse plasticity processes, we have extended the Arbor library to yield the Plastic Arbor framework, supporting simulations of a large variety of spike-driven plasticity paradigms (cf. Fig. 1b). To showcase the features of the new framework, we present examples of computational models, beginning with single-synapse dynamics [6,7], progressing to multi-synapse rules [8,9], and finally scaling up to large recurrent networks [10].
Results
While cross-validating our implementations by comparison with other simulators, we show that Arbor allows simulating plastic networks of multi-compartment neurons at nearly no additional cost in runtime compared to point-neuron simulations. Using the new framework, we have already been able to investigate the impact of dendritic structures on network dynamics across a timescale of several hours, showing a relation between the length of dendritic trees and the ability of the network to efficiently store information.
Discussion
Due to its modern computing architecture and inherent support of multi-compartment neurons, the Arbor simulator constitutes an important tool for the computational modeling of neuronal networks. By our extension of Arbor, we provide a valuable tool that will support future studies on the impact of synaptic plasticity, especially, in conjunction with neuronal morphology, in large networks. In our recent work, we also demonstrate new insights into the functional impact of morphological neuronal structure at the network level. In the future, the Plastic Arbor framework may power a great variety of studies considering synaptic mechanisms and their interactions with neuronal dynamics and morphologies, from single synapses to large networks.
Figure 1. Overview of the extended Arbor framework with support for synaptic plasticity simulations.
Acknowledgements
This work was supported by the German Research Foundation (Deutsche Forschungsgemeinschaft, DFG) through grants SFB1286 (C01, Z01) and TE 1172/7-1, as well as by the European Commission H2020 grants no. 899265 (ADOPD) and 945539 (HBP SGA3).
References1. https://doi.org/10.1109/EMPDP.2019.8671560
2. https://doi.org/10.1146/annurev.neuro.23.1.649
3. https://doi.org/10.1038/s41539-019-0048-y
4. https://doi.org/10.1016/j.conb.2008.08.013
5. https://doi.org/10.7554/eLife.46966
6. https://doi.org/10.1038/78829
7. https://doi.org/10.1073/pnas.1109359109
8. https://doi.org/10.1523/JNEUROSCI.0027-17.2017
9. https://doi.org/10.1371/journal.pone.0161679
10. https://doi.org/10.1038/s42003-021-01778-y

Speakers

Thorsten Hater

FZJ

Jannik Luboeinski

Christian Tetylaff

Sebastian Schmitt

Shirin Shafiee

Fabian Bösch

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P177: Cobrawap: from a specific use-case to a more general scientifically-technologically co-designed tool for neuroscience

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P177 Cobrawap: from a specific use-case to a more general scientifically-technologically co-designed tool for neuroscience

Cosimo Lupo1,*, Robin Gutzen2, Federico Marmoreo1, Alessandra Cardinale1,3, Michael Denker4, Pier Stanislao Paolucci1, Giulia De Bonis1

1Istituto Nazionale di Fisica Nucleare (INFN), Sezione di Roma, Rome, Italy
2Dept. of Psychology and Center for Data Science, New York University, New York, USA
3Università Campus Bio-Medico di Roma, Rome, Italy
4Institute for Advanced Simulation (IAS-6), Jülich Research Centre, Jülich, Germany

*Email:cosimo.lupo89@gmail.com
Introduction

Cobrawap (Collaborative Brain Wave Analysis Pipeline) [1-3] is an open-source, modular and customizable data analysis tool designed and implemented by INFN (Italy) and Jülich Research Centre (Germany) in the context of the Human Brain Project, further enhanced within the EBRAINS and EBRAINS-Italy initiatives. Its foundational goal was to enable standardized quantitative descriptions of cortical wave dynamics observed in heterogeneous data sources, both experimental and simulated, also allowing for validation and calibration of brain simulation models (Fig. 1). The current directions of development aim at enhancing generalizability beyond the set of originally considered use cases.

Methods
Intercepting the increasing demand by the Neuroscience community for reusability and reproducibility, Cobrawap provides a framework suitable for collecting generalized implementations of established methods and algorithms. Inspired by FAIR principles and leveraging the latest software solutions, Cobrawap is structured as a collection of modular Python3 building blocks that can be flexibly arranged along sequential stages, implementing data processing steps and analysis methods, directed by workflow managers (Snakemake or CWL). The collaborative approach behind the whole software allows users to seamlessly enrich its scope, by co-designing and implementing new processing or visualization blocks with the support of the Cobrawap “core team”.

Results
Cobrawap has been successfully appliedon murine data and data-driven simulations, for multi-scale quantitative comparisons of heterogeneous experimental datasets [4] and for validation and calibration of simulation models [5], in the specific use-case of cortical slow wave data analysis in low-consciousness brain states. Later applications on non-human primate experimental data, and on increasing levels of consciousness, have proven the robustness and the versatility of the approach, paving the way to the crucial extension toward human data. A fundamental step is represented by the comparison with simulations, e.g. via TheVirtualBrain [6,7], which allow both to benchmark the new algorithms, and to validate and calibrate such models [8,9].

Discussion
Cobrawap has proven to be effective in the analysis of both synthetic and experimental data of different origin, representing a FAIR-compliant collaborative framework for the scientific and technological co-design. Together with the appealing extension to experimental human data, in both physiological and pathological conditions, further lines of enhancement involve the analysis of the output from a variety of theoretical models, also including the outcomes of artificial neural networks; this makes it eligible for addressing the explainability of AI solutions in bio-inspired systems that incorporate the emulation of brain states as a key element for the implementation of efficient incremental learning and cognition [10,11].

Figure 1. Cobrawap offers standardized quantitative descriptions of brain wave dynamics observed in heterogeneous data sources, both experimental and simulated (top right panel), via a set of sequential stages featuring modular and flexible sets of processing and visualization blocks (bottom panel, for two different recording techniques on anesthetized mice), each easily customizable by the user.
Acknowledgements
Research co-funded by: European Union’s Horizon Europe Programme under Specific Grant Agreement No. 101147319 (EBRAINS 2.0); European Commission NextGeneration EU through Italian Grant MUR-CUP-B51E22000150006 EBRAINS-Italy PNRR.
References
[1]github.com/NeuralEnsemble/cobrawap
[2]cobrawap.readthedocs.io
[3]doi.org/10.5281/zenodo.10198748
[4] Gutzen, et al. (2024)doi.org/10.1016/j.crmeth.2023.100681
[5] Capone, De Luca, et al. (2023)doi.org/10.1038/s42003-023-04580-0
[6] Sanz Leon, et al. (2013)doi.org/10.3389/fninf.2013.00010
[7]www.thevirtualbrain.org
[8] Gaglioti, et al. (2024)doi.org/10.3390/app14020890
[9] Cardinale, Gaglioti, et al. (2025)in preparation
[10] Capone, et al. (2019)doi.org/10.1038/s41598-019-45525-0
[11] Golosio, De Luca, et al. (2021)doi.org/10.1371/journal.pcbi.1009045

Speakers

Cosimo Lupo

Robin Gutzen

Federico Marmoreo

Alessandra Cardinale

Michael Denker

Pier Stanislao Paolucci

Giulia De Bonis

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P178: Coupling brain network simulation with pharmacokinetics for Parkinson's disease: towards patient-usable digital twins

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

Sahar Moghimi

Fabrice Wallois

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P182: Computational aspects of microarousals during awakening from anesthesia

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P182 Computational aspects of microarousals during awakening from anesthesia

Arnau Manasanch*1,2, Leonardo Dalla Porta1, Melody Torao-Angosto1, Maria V. Sanchez-Vives1,3
1Institut d’Investigacions Biomèdiques August Pi i Sunyer (IDIBAPS), 08036 Barcelona, Spain
2Facultat de Medicina I Ciències de la Salut, Universitat de Barcelona, 08036 Barcelona, Spain
3ICREA, Passeig Lluís Companys 23, 08010 Barcelona, Spain

*Email:manasanch@clinic.cat

Introduction
The study of brain states is fundamental to understanding consciousness and its neural mechanisms [1,2]. Both sleep and anesthesia provide valuable models for investigating and characterizing brain states and their transitions [3,4,5]. While extensive research has characterized Microarousals (MAs), brief wake-like periods of brain activity, in sleep [6], these remain almost unexplored during anesthesia. Emerging evidence suggests that these transient events may be modulated by an infraslow rhythm [7,8,9], influencing arousal dynamics during emergence from anesthesia. Here, we investigate the dynamics of MAs during anesthetic emergence using local field potential (LFP) recordings from anesthetized rats, shedding light on infraslow modulation of transient arousals.

Methods

To obtain long-term LFP recordings in freely moving Lister-Hooded rats (6–10 months old), electrodes were chronically implanted 600 µm deep in the cortex. EMG was recorded from the neck muscle. After post-surgical care, animals underwent five days of handling before recordings. LFPs were recorded during anesthesia induction and emergence. The protocol was the same used in [10]. Briefly,each subject received a single shot of intraperitoneal anesthesia consisting of ketamine (20-40 mg/kg) and medetomidine (0.15-0.3 mg/kg).Cortical activity was monitored from wakefulness to full emergence from anesthesia. Experiments followed Spanish and EU regulations and were approved by the Ethics Committee of the Universitat de Barcelona (287/17 P3).
Results
After remaining in the slow oscillatory state, characterized by alternating Up (high firing) and Down (silent) periods, for 2–3 hours, the brain dynamics abruptly transitioned to a state dominated by fast oscillations (~6 Hz) and wake-like microarousals. As anesthesia wore off, MAs progressively increased in duration. This transition appeared to be modulated by an infraslow oscillation during a steady state period (~0.14 Hz), which gradually slowed (reaching ~0.04 Hz) in the progression towards wakefulness. Analysis of MAs across subjects reveals a consistent trend of increasing duration of the microarousals over time, with power-law distributions observed in the duration of MAs. These distributions show an average exponent of 2.33±0.36, suggesting that microarousals exhibit characteristic scaling behavior across different subjects.

Discussion
Our findings suggest that the increasing duration of microarousals (MAs) as anesthesia progresses reflect a gradual transition toward wakefulness, with dynamics that share properties with sleep awakening.The power-law behavior in MA duration indicates a scale-invariant process, a hallmark of self-organized criticality. This model provides a new understanding of the microarchitecture of anesthesia, offering a window into controlled microarousals and the network dynamics from unconsciousness to consciousness.

Acknowledgements
The EU Horizon 2020 Framework Programme for Research and Innovation under the Specific Grant Agreement No. 945539 (Human Brain Project SGA3); INFRASLOW PID2023-152918OB-I00 funded by MICIU / AEI / 10.13039/501100011033/FEDER. Co-funded by Departament de Recerca i Universitats de la Generalitat de Catalunya (AGAUR 2021-SGR-01165). IDIBAPS is funded by the CERCA program (Generalitat de Catalunya).
References
[1]https://doi.org/10.1038/nrn3084
[2]https://doi.org/10.1016/j.tins.2023.04.001
[3]https://doi.org/10.1126/science.8235588

[4]https://doi.org/10.1213/ANE.0000000000005361
[5]https://doi.org/10.1016/j.conb.2017.04.011
[6]https://doi.org/10.1016/s0987-7053(99)80016-1
[7]https://doi.org/10.1016/j.celrep.2021.109270
[8]https://doi.org/10.1016/j.neuron.2024.12.009
[9]https://doi.org/10.1038/s41593-024-01822-0
[10]https://doi.org/10.3389/fnsys.2021.609645

Speakers

Maria V. Sanchez-Vives

Leonardo Dalla Porta

Arnau Manasanch

Melody Torao-Angosto

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P183: Biophysically detailed neuron models with genetically defined ion channels

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P183 Biophysically detailed neuron models with genetically defined ion channels

DarshanMandge*1,2, Rajnish Ranjan2, Emmanuelle Logette2, Tanguy Damart2, Aurélien Tristan Jaquier2, Lida Kanari2, Daniel Keller2, Yann Roussel2, Stijn van Dorp2, Werner Van Geit1, and Henry Markram2
1Open Brain Institute,1005,Lausanne,Switzerland
2Blue Brain Project, Écolepolytechniquefédéralede Lausanne (EPFL), Campus Biotech, 1202 Geneva, Switzerland
*Email: darshan.mandge@openbraininsititute.org
Introduction

Cortical neurons can be classified into different electrical firing types (e-types). A common approachtomodellingthesee-types involvesthecreationofdetailed electrical models (e-models) using generic ion channel currents such as transient and persistent sodium, potassium channels, and high- and low-voltage activated calcium channels[1]. While this approach accurately captures a neuron's electricalbehaviour, it does notestablisha link between specificion channels andobservedelectrophysiological properties.
Methods
We now havemade47 homomeric ion channel modelscorresponding to various potassium[2], sodium, calcium, and hyperpolarization-activated cyclic nucleotide-gated (HCN) ion channels. These genetic ion channel models were based on independent experimental data from the heterologous expression of the corresponding genes. The genetic channels along with some generic ion channelswereused in this study to construct cortical e-type modelsfromdetailed morphological reconstructions and electrophysiological datacollected inthe rat somatosensory cortex. Webuilt aPython-basedpipeline calledBluePyEModel[3] tobuild such e-models.
Results
The optimized e-modelsreproduce firing propertiesobservedinin vitrorecordings. Electrical features of the optimized e-models were found to be within 3–5 standard deviations of the corresponding mean experimental recordings.
Discussion
These biophysically detailed modelsenablea better understanding of the electrical activity in normal and pathological states of neurons.Inthefuture, wewillmake these e-models available on the Open Brain Institute (OBI)platform,https://openbraininstitute.org/. The OBIplatformprovidesa comprehensive repository of digital brain models andstandardisedcomputational modelling services to enableusers to conduct realistic brain simulations, test hypotheses, and explore the complexitiesatvarious modelling levels– Subcellular, Cellular, Circuit andSystems.

Acknowledgements
This study was supported by funding to the Blue Brain Project, a researchcenterof the Écolepolytechniquefédéralede Lausanne (EPFL), from the Swiss government’s ETH Board of the Swiss Federal Institutes of Technology.

References
1.https://doi.org/10.1016/j.patter.2023.100855
2.https://doi.org/10/ghqvg8
3.https://doi.org/10.5281/zenodo.8283490

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P184: Spontaneous oscillations and neural avalanches are linked to whisker stimulation response in the rat-barrel and thalamus circuit

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P184 Spontaneous oscillations and neural avalanches are linked to whisker stimulation response in the rat-barrel and thalamus circuit

Benedetta Mariani*1, Ramon Guevara Erra1,2, Mattia Tambaro3,4, Marta Maschietto3, Alessandro Leparulo3,5, Stefano Vassanelli3, Samir Suweis1,2
1Padova Neuroscience Center, University of Padova, Padova, Italy
2Department of Physics and Astronomy, University of Padova, Padova, Italy
3Department of Biomedical Sciences, University of Padova, Padova, Italy
4Department of Physics, University of Milano Bicocca, Milan, Italy
5Department of Neuroscience, University of Padova, Padova, Italy

*Email: benedetta.mariani@unipd.it

Introduction
The cerebral cortex operates in a state of restless activity, even in the absence of external stimuli [1,2]. Collective neuronal activities, such as neural avalanches[3] and collective oscillations[4], are also found under resting conditions, and these features have been suggested to support sensory processing and brain readiness for rapid responses [2]. However, most of these results are supported by theoretical models rather than experimental observations. The rat barrel cortex and thalamus circuit, with its somatotopic organization for processing whisker movements, provides a powerful system to explore the interplay between spontaneous and evoked activities.
Methods
To characterize the resting state circuits, we perform multi-electrode recordings in both rats' barrel cortex and thalamus through a neural probe, both during spontaneous activity and activity after controlled whisker stimulation. We decompose the LFP signals into their frequency contents through Empirical Mode Decomposition, a tool that is suited to analyze non-linear and non-stationary oscillations. We also analyze avalanches distributions by detecting events in MUAs activity and grouping them by temporal proximity. We then employ a mesoscopic firing rate model, fitted on real data [5], to understand the observed phenomomenology. It receives as input the experimental thalamic firing rate.

Results
During spontaneous activity, we find 10-15 Hz oscillations in the barrel cortex concomitantly with slow 1-4 Hz oscillations, as well as power-law distributed avalanches. The slow oscillations are also present in the thalamus, while the 10-15 Hz one is lacking. We find that the phase of the slow oscillation modulates the higher frequency amplitude, as well as avalanche occurrences. We then record neural activity during controlled whisker movements to confirm that the 10-15 Hz barrel circuit is amplified after whisker stimulation.We finally show how the thalamic-driven firing rate model can describe the entire phenomenology observed and predict the response to whisker stimulations.
Discussion

Our results show that even during spontaneous activity the rat barrel cortex displays a rich dynamical state that includes avalanches and oscillations, which are coupled through the slow oscillation. The 10-15 Hz oscillation is amplified after the whisker stimulation, suggesting that spontaneous neural activity primes the rat cortex for the whisker response. These facts are confirmed by our model, that is able to reproduce the resting state phenomenology and the amplification of oscillations after stimulation, thanks to the thalamic input to the cortex. Moreover, the barrel cortex oscillatory behavior may allow a flexible synchronization mechanism for the perception of stimuli.

Acknowledgements
Work by B.M and S.S. is supported by #NEXTGENERATIONEU (NGEU) and funded by the Ministry of University and Research (MUR), National Recovery and Resilience Plan (NRRP), project MNESYS (PE0000006) –(DN. 1553 11.10.2022). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

References
[1] Raichle M. E. (2011),https://doi.org/10.1089/brain.2011.0019
[2]Smith, S. M., et al (2009),https://doi.org/10.1073/pnas.0905267106
[3]Beggs, J. M., & Plenz, D. (2003),https://doi.org/10.1523/JNEUROSCI.23-35-11167.2003
[4] Singer W. (2018),https://doi.org/10.1111/ejn.13796
[5]Pinto, D. et al (1996),https://doi.org/10.1007/BF00161134

Speakers

Benedetta Mariani

Ramon Guevara Erra

Mattia Tambaro

Marta Maschietto

Alessandro Leparulo

Speakers

Egidio D'Angelo

Stefano Masoli

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P188: Neural coding of subthreshold sinusoidal inputs into symbolic temporal spike patterns

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P188 Neural coding of subthreshold sinusoidal inputs into symbolic temporal spike patterns

Maria Masoliver1,Cristina Masoller*1
1Departmento de Física, Universitat Politecnica de Catalunya, Terrassa, Spain
*Email: cristina.masoller@upc.edu

Introduction

Neuromorphic photonics is a new paradigm for optical computing that can revolutionize the fields of signal processing and artificial intelligence.To develop photonic neurons able to process information as sensory neurons do, we need to identify excitable lasers able to emit pulses of light (optical spikes) that similar to neuronal spikes, and implement in these lasers the neural coding mechanisms used by neural systems to process information, in particular, the neural coding mechanisms used to process weak external inputs in noisy environments.
Methods
We use thestochastic FitzHugh Nagumo model to simulate spike sequences fired in response to weak (subthreshold) sinusoidal signals.We also use this model to simulate the activity of a population of neurons, when they all perceive the same subthreshold sinusoidal input. We use a symbolic time series analysis method, known as ordinal analysis [1], to analyze the sequences of inter-spike-intervals.

Results

In the analysis of the spikes of single neurons, we found that the probabilities of the symbols (ordinal patterns) encode information of the signal, because they depend on both, the amplitude and the frequency of the signal.
In the analysis of the spike generated by a population of neurons, we have also found that the ordinal probabilities encode information of the amplitude and of the period of the signal that is perceived by the neurons. We found that neuronal coupling benefits signal encoding because groups of neurons are able to encode a small-amplitude signal that cannot be encoded when it is perceived by just one or two neurons. Interestingly, we found that for a population of neurons, just a few random links between them can significantly improve signal encoding.
Discussion

We have found that the probabilities of spike patterns in spike sequences may encode information of a weak (subthreshold) input perceived by the neurons.
An open question is whether this coding mechanism can be implemented in excitable lasers that emit pulses of light (optical spikes) whose statistical properties are similar to neuronal spikes.Using ordinal analysis and machine learning, we have found that the sequences of optical spikes emitted by a laser diode in response to low or high frequency signals are located in different regions of a 3D feature space, suggesting that information about the frequency of the input signal, can be recovered from the analysis of the emitted optical spikes [3].

Figure 1. Left: Optical spikes emitted by an excitable laser (nanosecond time scale); right: neuronal spikes simulated with the FitzHugh Nagumo model (millisecond time scale).
Acknowledgements
Ministerio de Ciencia, Innovación y Universidades (No. PID2021-123994NB-C21), Institució Catalana de Recerca i Estudis Avançats (ICREA Academia), Agencia de Gestió d’Ajuts Universitaris i de Recerca (AGAUR, No. 2021 SGR 00606).
References
[1] Bandt C, Pompe B. (2002). Permutation entropy: a natural complexity measure for time series. Phys. Rev. Lett., 88, 174102.https://doi.org/10.1103/PhysRevLett.88.174102

[2] Masoliver M, Masoller C (2020). Neuronal coupling benefits the encoding of weak periodic signals in symbolic spike patterns. Commun. Nonlinear Sci. Numer. Simulat. 88, 105023.https://doi.org/10.1016/j.cnsns.2019.105023

[3] Boaretto BRR, Macau EEN, Masoller C (2024). Characterizing the spike timing of a chaotic laser by using ordinal analysis and machine learning, Chaos 34, 043108.https://doi.org/10.1063/5.0193967

Speakers

Cristina Masoller

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P189: Elementary Dynamics of Neural Microcircuits

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P189 Elementary Dynamics of Neural Microcircuits

Stefano Masserini*1,2,3, Richard Kempter1,2,3

1 Institute for Theoretical Biology, Humboldt-Universität zu Berlin, Berlin, Germany
2Bernstein Center for Computational Neuroscience, Berlin, Germany
3Charité-Universitätsmedizin Berlin, Einstein Center for Neurosciences, Berlin, Germany

*Email: stefanomasse@gmail.com

Introduction

Cell type diversity is a major direction in which systems neuroscience has expanded in the last decade, as networks of excitatory (E) and inhibitory (I) neurons enriched with specific neuronal populations with their own distinct role in the network dynamics. These advances have mostly been driven by new experimental techniques, often inspiring circuit-specific modeling, even when stark similarities across cortical areas would have allowed describing the dynamics of these microcircuits with a more general mathematical language. Steps toward a general description have been taken by using linear approximations to understand how connectivity shapes responses to perturbations from within or outside the network [1,2].

Methods
In this work, we expand on these findings by studying microcircuit dynamics in the simplest nonlinear model, the threshold-linear network (TLN), and generalize insights originally obtained for all-inhibitory TLNs [3]. This model greatly extends the dynamical repertoire of purely linear networks, by allowing for oscillations and multistability. On the other hand, it retains the simplicity of linear models, since the conditions for each nonlinear regime can be computed in closed form and intuitively interpreted in terms of input and connectivity requirements. With this tool, we not only map previously unrelated systems neuroscience hypotheses to a common reference space, but also gain new insights into specific circuits across the brain.
Results
Namely, we compare balancing strategies in inhibition-stabilized E-I networks and discuss different types of bistability in hippocampal E-I-I networks. We then examine the conditions for gamma oscillations in the canonical circuit (Fig. 1A), providing a mechanistic explanation for the opposing effects of PV and SOM interneurons [4]. In E-E-I circuits, we show that connectivity determines three fundamentally different types of assembly interactions, while in E-E-I-I circuits we find that balanced clustering prevents coordinated inputs to one E-I unit from exerting lateral inhibition (Fig. 1B), while opponent clustering can induce competition even between strongly coupled E assemblies, resulting in different bistable configurations (Fig. 1C).
Discussion
While TLNs have so far not been regarded as a standard rate model for neural populations, these applications show that they can provide interpretable conditions even for the emergence of complex dynamical landscapes. These conditions should be taken into account by future modeling work on neural microcircuits, at least as a benchmark to determine whether additional complexity is necessary to explain their dynamics of interest. The simple structure of this model is also amenable to the addition of variables representing synaptic plasticity or slow adaptive currents. TLNs can also be directly compared to spiking networks, for example because they are the first-order mean-field limit for networks of Poisson neurons [5].

Figure 1. (A) Canonical circuit. (Aii-iii) Oscillation coherence. (Aiv) Effects of impairing SOM or PV (matching shading). (B-C) EEII network. (Bii-iii) Firing modulation wrt bottom left point. (Biv) Modulation example (shaded area). (Cii) Dynamical landscape, smaller regions are EII or EEI bistability. (Ciii) Lateral inhibition by either inputs to E1 or I1. (Civ) EI bistability. Input to I1 induces switch.
Acknowledgements
The authors thank Gaspar Cano, Carina Curto, Atilla Kelemen, John Rinzel, Archili Sakevarashvili, and Tilo Schwalger for insightful discussions about this study. Founding source: German Research Foundation, project 327654276--SFB 1315.
References
[1]https://doi.org/10.1101/2020.10.13.336727
[2] https://doi.org/10.1073/pnas.231104012
[3] https://doi.org/10.48550/arXiv.1804.00794
[4]https://doi.org/10.1038/nn.4562
[5] https://doi.org/10.48550/arXiv.2412.16111

Speakers

Richard Kempter

Stefano Masserini

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P190: Simulating healthy and diseased behaviour using spiking neurons

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P190 Simulating healthy and diseased behaviour using spiking neurons

Mavritsaki E.1,3, Klein, J.1, Porwal, N.1, Allen, H.A.2, Bowman, H.3, Amanatidou, V.4, Cook, A.1, Clibbens, J.1and Lintern, M.1

1College of Psychology, Birmingham City University, Birmingham, UK
2School of Psychology, University of Nottingham, Nottingham, UK
3School of Psychology, University of Birmingham, Birmingham, UK
4Worcestershire Health and Care Trust, UK

*Email: eirini.mavritsaki@bcu.ac.uk

Introduction

Spiking neural networks have proven highly effective in simulating both healthy and diseased neural behaviour. They offer researchers the opportunity to simultaneously study behaviour and understand its relationship with the underlying biological properties of the system. The approach is particularly valuable as these networks accurately mimic real neuronal communication, providing a more biologically accurate model compared to traditional methods, allowing researchers to analyse time-dependent patterns and providing deeper insights into neural dynamics and cognitive processes. Consequently, spiking neural networks have become an invaluable tool for advancing brain studies and neurological research. In this work, we present two studies utilizing spiking neural networks extending our previous work using the spiking Search over Time and Space (sSoTS) model.
Methods
The sSoTS model is a spiking neural model incorporating a fast excitatory AMPA recurrent current, a slow excitatory NMDA current, an inhibitory GABA current, and aIAHPslow[Ca+]activatedK+current.We built upon our previous research in visual search (Mavritsaki et al., 2011; Mavritsaki & Humphreys, 2016) to simulate behavioural findings in our lab in attention between adults, children and children that score high in Conners 3AI, testing ADHD. We also build upon our previous Alzheimer’s work (Mavritsaki et al., 2019) to simulate N400 and P600 components in the semantic category judgment task (Olichney et al., 2000), which has been used to track ERP changes in patients progressing through MCI to mild AD. Please see figure 1.

Results
Results from our visual search paradigm demonstrate that reducing coupling between neurons in the model successfully simulates the differences between adults and children. Furthermore, our findings suggest that temporal binding between feature items may be a key mechanism underlying differences observed between healthy children and those scoring high on the Conners 3AI test, as reducing this parameter in the model reproduced the observed differences. In our Alzheimer's work, we simulated the biomarkers found with the N400 and P600 ERP components by modelling the semantic category judgment task and modifying parameters related to pathological ionic, neurotransmitter, and atrophy modulations.

Discussion
Results from both studies demonstrate the importance of using spiking neural networks in computational modelling, as they provide valuable insights into brain functions, link different methodologies, and help understand changes that occur in diseased brains. Our Alzheimer's work shows that the disease's pathology can be measured through N400 and P600 congruency effects, thus validating ERPs as biomarkers for AD. Our visual search and ADHD work identifies the crucial role of binding in visual search and provides valuable insights into the ADHD condition that can support updates to the diagnostic criteria for ADHD.

Figure 1. The top part of the figure illustrates the key neuronal properties of the spiking neural network model. The bottom left panel shows the network connectivity implemented to simulate the semantic category judgment task in our Alzheimer's disease study, while the bottom right panel depicts the neural network configuration used to simulate visual search task performance in our ADHD behavioural study.
Acknowledgements
The computations described in this paper were performed using the University of Birmingham's BEAR Cloud service, which provides flexible resource for intensive computational work to the University's research community. Seehttp://www.birmingham.ac.uk/bearfor more details.
References
Mavritsaki, E., Bowman, H., & Su, L. (2019). Springer Intern.Publishing. https://doi.org/10.1007/978-3-030-18830-6_11
Mavritsaki, E., Heinke, D., Allen, H., Deco, G., & Humphreys, G. W. (2011). Bridging the Gap Between Physiology and Behavior:Psych.l Review,118(1), 3–41. https://doi.org/10.1037/A0021868
Mavritsaki, E., & Humphreys, G. (2016).Journal of Cognitive Neuroscience,28(10). https://doi.org/10.1162/jocn_a_00984

Olichney, J. M., Van Petten, C., Paller, K. A., Salmon, D. P., Iragui, V. J., & Kutas, M. (2000).Brain,123(9), 1948–1963. https://doi.org/10.1093/brain/123.9.1948

Speakers

Eirini Mavritsaki

Jennifer Klein

Nirav Porwal

Harreit Allen

Howard Bowman

Virginia Amanatidou

Amy Cook

John Clibbens

Maxine Lintern

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P191: "Brain Fluidity as a Biomarker for Alzheimer's Disease: Linking Network Dynamics to Clinical Disability Prediction"

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P191 "Brain Fluidity as a Biomarker for Alzheimer's Disease: Linking Network Dynamics to Clinical Disability Prediction"

Camille Mazzara*1,2,3, Gian Marco Duma4, Giuditta Gambino5, Giuseppe Giglia5, Michele Migliore2, Pierpaolo Sorrentino3,6,7
1.Department of Promoting Health, Maternal-Infant. Excellence and Internal and Specialized Medicine (PROMISE) G. D’Alessandro, University of Palermo, Palermo, Italy.
2.Institute of Biophysics, National Research Council, Palermo, Italy.
3.Aix Marseille Univ, INSERM, INS, Inst Neurosci Syst, Marseille, France.
4.IRCCS E. Medea Scientific Institute, Conegliano, Treviso, Italy
5.Department of Biomedicine, Neuroscience and Advanced Diagnostics (BIND), University of Palermo, Palermo, Italy.
6.Institute of Applied Sciences and Intelligent Systems, National Research Council, Pozzuoli, Italy
7.University of Sassari, Department of Biomedical Sciences, Viale San Pietro, 07100, Sassari, Italy.
*email: camille.mazzara@ibf.cnr.it

Introduction

Alzheimer’s disease (AD) is a neurodegenerative disorder characterized by progressive cognitive decline and large-scale network dysfunction[1]. While amyloid-beta (Aβ) and tau pathology are well-documented [2,3], the network-level mechanisms linking neuronal degeneration to cognitive impairment remain poorly understood. Traditional functional connectivity (FC) analyses provide static representations of brain networks, failing to capture their intrinsic dynamics[4]. We proposebrain fluidity, a metric that quantify network flexibility, as a potential biomarker reflecting AD-related disruptions in brain dynamics.

Methods
After preprocessing and source reconstruction, we analyzed resting-state EEG data from 28 AD patients and 29 healthy controls. Brain fluidity was quantified measuring the variability of functional connectivity over time, reflecting how interregional synchronization evolves. We assessed its relationship with established AD biomarkers, including cerebrospinal fluid (CSF) levels of Aβ42, phosphorylated tau (p-tau), and total tau (t-tau). Additionally, we examined associations between brain fluidity and cognitive performance (Mini-Mental State Examination (MMSE)). Statistical analyses included between-group comparisons and regression models to determine the predictive value of fluidity in tracking disease severity.
Results
Fluidity analysis across frequency bands (theta, alpha, beta, gamma) revealed significant differences in AD patients (Fig.1a). In the theta band (4–8 Hz), fluidity was higher in AD compared to controls, while in the beta band (14–30 Hz), fluidity was lower. Correlation analyses showed no significant associations between theta fluidity and clinical measures. However, beta fluidity negatively correlated with tTau and pTau (Fig.1c), suggesting a link to neurodegeneration. Notably, no significant associations were found between fluidity and Aβ levels. Using a multilinear regression model we also found that adding fluidity calculated in the beta band significantly improved the predictive power for clinical disability.
Discussion
This results could imply that changes in the ability of the brain to flexibly switch between different dynamic states are associated with neurodegenerative processes, specifically tau-related damage. Reduced brain fluidity in beta may reflect underlying neurodegenerative processes, providing insights into the functional consequences of neuronal loss. Given its sensitivity to AD-related changes, brain fluidity may serve as a promising biomarker for tracking disease progression and evaluating treatment efficacy in clinical settings.

Figure 1. Fig.1 a) Fluidity for each frequency band in AD and control groups. b) dFC matrices averaged across AD (left) and control (right), computed in theta (top) and beta (bottom). c) Correlation between beta-band fluidity and tTau (left), pTau (center), Aβ (right), with significant links to tTau (p = 0.03) and pTau (p = 0.01), but not Aβ42.
Acknowledgements

References
Bibliography
1.https://doi.org/10.1016/j.lfs.2020.117996
2.https://doi.org/10.1590/S1980-57642009DN30300003

3.https://doi.org/10.7554/eLife.98920.1

Speakers

Camille Mazzara

Michele Migliore

Pierpaolo Sorrentino

Gian Marco Duma

Giuditta Gambino

Giuseppe Giglia

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P192: Identifying Cell-Type-Specific Alterations Underlying Schizophrenia-Related EEG Deficits Using a Multiscale Model of Auditory Thalamocortical Circuits

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P192 Identifying Cell-Type-Specific Alterations Underlying Schizophrenia-Related EEG Deficits Using a Multiscale Model of Auditory Thalamocortical Circuits

Scott McElroy*1,2, James Chen1,2, Nikita Novikov1,2,3, Pablo Fernández-López4, Carmen Paz Suárez-Araújo4, Christoph Metzner5, Daniel Javitt3, Sam Neymotin3, Salvador Dura-Bernal1,2,3
1Global Center for AI, Society and Mental Health, SUNY Downstate Health Sciences University, Brooklyn, United States of America
2Department of Physiology and Pharmacology,SUNY Downstate Health Sciences University, Brooklyn, United States of America
3Center for Biomedical Imaging & Neuromodulation, Nathan Kline Institute, Orangeburg, United States of America
4Instituto Universitario de Cibernética, Empresa y Sociedad, Universidad de Las Palmas de Gran Canaria, Gran Canaria, España
5Technische Universität Berlin, Berlin, Germany

*Email: scott.mcelroy@downstate.edu
Introduction
Schizophrenia is associated with cognitive deficits, including disruptions in sensory processing. Electroencephalography (EEG) studies have identified abnormalities in event-related potentials and cortical oscillations, particularly within the auditory system. Among the most well-established EEG biomarkers are the reduced 40 Hz Auditory Steady-State Response (ASSR) and impaired mismatch negativity (MMN). Understanding the neural mechanisms underlying these EEG deficits is critical for linking molecular and circuit-level alterations to cognitive dysfunctions in schizophrenia.Methods
We extended our computational model of auditory thalamocortical circuits to investigate the circuit-level mechanisms underlying schizophrenia-related EEG abnormalities1. The model simulates a cortical column with over 12,000 neurons and 30 million synapses, incorporating experimentally derived neuron densities, laminar organization, morphology, biophysics, and connectivity across multiple scales. Auditory inputs to the thalamus were modeled using a phenomenological cochlear representation, allowing for the reproduction of realistic physiological responses. Additionally, a more systematic approach to providing background network activity was implemented using Ornstein-Uhlenbeck (OU) processes to model time-varying, statistically independent somatic conductance injections.Results & Discussion
Our refinements enhance the physiological fidelity of EEG simulations, enabling improved replication of schizophrenia-related biomarkers. The integration of OU-modeled background activity ensures smoother, correlated variations in network input, leading to more biologically realistic fluctuations in neuronal dynamics. The OU process's mean and standard deviation are expressed as input conductance percentages for each cell type, linking them to intrinsic cellular properties. Additionally, we are developing an adaptive algorithm to dynamically calibrate population-specific OU parameters, ensuring model flexibility as it evolves. By incorporating experimentally observed molecular and genetic alterations, our model provides deeper insights into the neural basis of auditory processing deficits in schizophrenia and strengthens the link between cellular dysfunctions and EEG biomarkers.

Acknowledgements
This work is supported by NIBIB U24EB028998
References
1.10.1016/j.celrep.2023.113378

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P196: Kernel-based LFP estimation in detailed large-scale spiking network model of visual cortex

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P196 Kernel-based LFP estimation in detailed large-scale spiking network model of visual cortex

Nicolò Meneghetti1,2,*, Atle E. Rimehaug3, Gaute T. Einevoll4,5, Alberto Mazzoni1,2, Torbjørn V. Ness4

1The Biorobotics Institute, Scuola Superiore Sant’Anna, Pisa, Italy
2Department of Excellence for Robotics and AI, Scuola Superiore Sant’Anna, Pisa, Italy
3Department of Informatics, University of Oslo, Oslo, Norway
4Department of Physics, Norwegian University of Life Sciences, Ås, Norway
5Department of Physics, University of Oslo, Oslo, Norway

*Email: nicolo.meneghetti@santannapisa.it

Introduction

Large-scale neuronal networks are fundamental tools in computational neuroscience. A key challenge in this domain is simulating measurable signals like local field potentials (LFPs), which bridge the gap between in silico model predictions and experimental data. Simulating LFPs in large-scale models, however, requires biologically detailed multicompartmental (MC) neuron models, which impose significant computational demands. To address this, multiple simplified approaches have been developed. In our work [1] we extended a kernel-based method to enable accurate LFP estimation in a state-of-the-art MC model of the mouse primary visual cortex (V1) from the Allen Institute [2], [3] while significantly reducing computational costs.

Methods
This V1 model features extensive biological detail, with over 50,000 MC neurons across six cortical layers[2], as well as experimentally recorded afferent inputs from both thalamic and lateromedial visual areas[3].
Instead of direct MC simulations, our method estimates the LFP by convolving population firing rates with precomputed spatiotemporal kernels (Fig. 1A), which represent the average postsynaptic LFP response to a presynaptic spike (see e.g., Fig.1B). This drastically reduced computational cost while maintaining estimation accuracy.
Results
The kernel method accurately estimated LFPs in both superficial (Fig. 1C) and deep layers (Fig. 1D). By treating LFPs as the sum of convolutions of neuronal firing rates and LFP-kernels, the method also enabled disentangling the contributions of different neuronal populations to the overall LFP. We found that V1 LFPs are primarily driven by external inputs, with thalamic afferents dominating in layer 4 (Fig. 1F) and lateromedial feedback influencing L2/3 layers (Fig. 1E). In contrast, local synaptic activity contributed minimally, challenging the conventional view that PV neurons are primary LFP drivers [4]. In fact, we showed that PV apparent influence on LFP reflects their correlation with external inputs rather than direct contribution.
Discussion
Our findings establish the kernel-based method as a robust and efficient tool for LFP estimation in large-scale network models. By significantly reducing computational costs, this approach makes detailed LFP simulations more practical while also providing insights into cortical LFP generation. Our results highlight the predominant role of external synaptic inputs, while challenging the conventional view that local network activity, including inhibitory interneurons, is a primary LFP driver. This methodology provides a useful framework for studying sensory processing and network dynamics in large-scale models, helping to clarify the contributions of different neuronal populations to cortical LFPs.

Figure 1. (A) Schematic of the kernel-based LFP estimation. (B) Set of kernels for computing L2/3 LFPs for different presynaptic families. (C) L2/3 LFPs computed with both MC simulations (red) and kernel convolution (black). (D) Same as C, for layer 4. (E) Cross-R² matrix between the total L2/3 LFPs and the LFP generated by the synaptic activity of each population in the model. (F) Same as C, for layer 4.
Acknowledgements
This work was supported by the Ministry of University and Research (MUR), National Recovery and Resilience Plan (NRRP), project MAD-2022-12376927 (“The etiopathological basis of gait derangement in Parkinson’s disease: decoding locomotor network dynamics”).
References
[1]https://doi.org/10.1101/2024.11.29.626029
[2]https://doi.org/10.1016/j.neuron.2020.01.040
[3]https://doi.org/10.7554/eLife.87169
[4]https://doi.org/10.1038/srep40211

Speakers

Alberto Mazzoni

Gaute Einevoll

Torbjørn V. Ness

Atle Rimehaug

Nicolo Meneghetti

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P197: Resilience of local microcircuitry firing dynamics to selective connectivity degeneration

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P197 Resilience of local microcircuitry firing dynamics to selective connectivity degeneration

Simachew Mengiste*1, Ad Aertsen2, Demian Battaglia1, Arvind Kumar3

1Functional System Dynamics / LNCA UMR 7364, University of Strasbourg, France
2BCCN / University of Freiburg, Germany
3KTH Royal Institute of Technology, Stockholm, Sweden

*Email: mengiste@unistra.fr

Introduction
Local microcircuit connectivity within local cortical microcircuits shapes spiking dynamics, influencing firing rate, synchrony, and regularity (thus information bandwidth). Often modeled as random and sparse (Erdös-Rényi, ER) or with small-world or scale-free properties, connectivity derived from detailed connectomic reconstructions (Egger et al., 2014) display dense cell clusters, diverging from mere randomness.
Neurodegenerative diseases (e.g., Alzheimer’s) induce neuronal and synaptic loss, disrupting dynamics. We systematically examine how pruning affects microcircuits with different topologies, revealing that resilience strongly depends on connectivity, with the "real connectome" being particularly robust.

Methods
We studied three random network topologies—Erdös-Rényi (ER), small-world (SW), and scale-free (SF)—plus a fourth based on real connectome (RC) reconstructions. Neurons were modeled as leaky integrate-and-fire units, with excitatory and inhibitory inputs shaping membrane potential dynamics.

Network degeneration was simulated via progressive pruning of synapses and neurons, using random or targeted sequences based on node degree or centrality. We analyzed firing rate, correlations, and spiking variability (coefficient of variation), alongside net synaptic currents received on average. Structural changes were assessed via graph metrics. We then systematically probed how firing dynamics evolved in the four ensembles along neurodegeneration.

Results
Using different network topologies and neurodegenerative strategies, we found that activity states were largely independent of topology across the different ensembles. Degeneration induced similar firing rate and synchrony variations across neurodegenerative schemes. We hypothesized that E-I balance changes, rather than topology, drove these dynamics. The effective synaptic weight (ESW) best predicted network activity, explaining firing rate, variability, and synchrony—except pairwise correlation, which depended on shared presynaptic neighbors and connection density. The real connectome (RC) followed similar ESW dependencies but exhibited broader stability ranges for all different firing parameters.

Discussion
While most neurodegeneration models focus on long-range connectivity changes, local microcircuits are also affected, altering synchrony and information processing. We find that local circuit dynamics are indeed disrupted, but less dependent on precise connectivity than expected. Instead, the effective synaptic weight (ESW) emerges as a stronger predictor of network behavior, making it a key measure for assessing function in both healthy and diseased states. The anomalous stability of firing parameters in networks with realistic connectivity suggests that microcircuit properties may have evolved to enhance functional resilience.

Acknowledgements
ANR PEPR Santé Numérique "BHT - Brain Health Trajectories"
ReferencesEgger, R., Dercksen, V. J., Udvary, D., Hege, H.-C., & Oberlaender, M. (2014). Generation of dense statistical connectomes from sparse morphological data.Frontiers in Neuroanatomy,8, 129. https://doi.org/10.3389/fnana.2014.00129

Speakers

Demian Battaglia

Arvind Kumar

Simachew Mengiste

Ad Aertsen

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P198: Binary Brains: Excitable Dynamics Simplify Neural Connectomes

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P198 Binary Brains: Excitable Dynamics Simplify Neural Connectomes

Arnaud Messé¹, Marc-Thorsten Hütt², Claus C. Hilgetag¹,³*
¹ Institute of Computational Neuroscience, Hamburg Center of Neuroscience, University Medical Center Eppendorf, Hamburg, Germany
² Computational Systems Biology, School of Science, Constructor University, Bremen, Germany
³ Department of Health Sciences, Boston University, Boston, MA, USA
--
*Email: c.hilgetag@uke.de

Introduction
Neural connectomes, representing the structural backbone of brain dynamics, are traditionally analyzed as weighted networks. However, the computational cost and methodological challenges of weighted representations hinder their widespread use. Here, we demonstrate that excitable dynamics—a common mechanism in neural and artificial networks—enable a thresholding approach that renders weighted and binary networks functionally equivalent. This finding simplifies network analyses and supports efficient artificial neural network (ANN) design by drastically reducing memory and computational demands.
Methods
We examined excitable network dynamics using a cellular automaton-based excitable model (SER model) and the FitzHugh-Nagumo model. By mapping the local excitation threshold onto global network weights, we identified a threshold at which binarized networks produce activity patterns statistically indistinguishable from those in weighted networks. Simulations were performed on synthetic networks, empirical structural brain connectivity data (MRI-derived), and artificial neural networks trained on the MNIST dataset. Computational efficiency was assessed in terms of memory usage and execution time.
Results & Discussion
Our findings [1] show that, under appropriate thresholding, binarized networks accurately reproduce coactivation patterns and functional connectivity observed in weighted brain networks. This effect holds across diverse network topologies and weight distributions, particularly for log-normal weight distributions found in empirical data. Computationally, binarized networks require significantly less memory and reduce processing times by orders of magnitude. These findings not only simplify empirical network analyses in neuroscience but also suggest a general principle for optimizing computational models in various domains, including machine learning, complex systems, and bio-inspired AI. Particularly in ANNs, thresholding maintains classification accuracy while drastically lowering the number of parameters, making binary networks a promising approach for efficient AI design.

Acknowledgements
The research was supported by the Deutsche Forschungsgemeinschaft (DFG) - SFB 936 - 178316478 - A1 & Z3, SPP2041 - 313856816 - HI1286/7-1, TRR 169 - 261402652 - A2, and the EU Horizon 2020 Framework Programme (HBP SGA2 & SGA3).
References
[1]https://doi.org/10.1101/2024.06.23.600265

Speakers

Claus Hilgetag

Arnaud Messé

Marc-Thorsten Hütt

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P199: Efficient slope detection with regular spiking and bursting point neuron models

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P199 Efficient slope detection with regular spiking and bursting point neuron models

Rebecca Miko*1, Marcus M. Scheunemann2,3, Volker Steuber1, Michael Schmuker1

1Biocomputation Research group, University of Hertfordshire, Hatfield, UK
2Adaptive Systems Research Group, University of Hertfordshire, Hatfield, UK
3Autonomy Department, Dexory, London, United Kingdom

*Email: rebeccamiko@outlook.com

Introduction

In real-world environments, odour stimuli exhibit a complex temporal structure due to turbulent gas dispersion, resulting in intermittent and sparse signals. These turbulence-induced fluctuations can be rapid yet contain valuable information crucial for locating odour sources. This ability is essential for both biological agents in foraging and mate-seeking behaviours, as well as robotic gas sensing in environmental and industrial monitoring. However, omnipresent turbulence destroys concentration gradients. Research suggests that the temporal dynamics of odour signals encode key information about the olfactory scene [1, 2].

Methods
Using the Izhikevich model [3], we develop neurons that spike at the rising edges (Fig. 1a) of naturalistic input signals across varying frequencies. We then compare these neurons to a two-compartmental model [4], which predominantly fires bursts at positive slopes in naturalistic inputs. By analysing the spiking behaviour of both models, we assessed whether bursting mechanisms are necessary for detecting odour signal dynamics
Results
Our findings indicate that a regular spiking neuron can effectively encode the slopes of input signals through discrete spike events, and that these detectors do not need to have the bursting mechanism (Fig. 1b). In contrast, the two-compartmental model [4] predominantly fires bursts in response to rising signal slopes, while the Izhikevich model generates single spikes at these transitions while maintaining computational efficiency. This demonstrates that a simple spiking neuron can capture key temporal features of odour signals without complex bursting dynamics.
Discussion
These results suggest that detecting odour signal slopes does not require burst firing. Instead, regular spiking neurons can efficiently encode temporal features of turbulent odour signals. Given the computational efficiency of the Izhikevich point neuron model, our findings offer potential applications in robotic gas navigation, where rapid and accurate data processing is crucial. By leveraging simple neural mechanisms, future research can explore bio-inspired gas-sensing systems for environmental and industrial monitoring.

Figure 1. Top trace: Gaussian white noise input nA (5 Hz; µ = .006; σ = .015). Bottom trace: membrane potential mV response. Top panel: neuron has parameters {a:0.01,b:0.2,c:- 35,d:5.0}. Asterisks mark burst onsets (grey dotted lines added for clarity). Bursts are defined by ISI ≤ 10 ms. No single spikes were produced. Bottom panel: neuron has parameters {a:0.01,b:0.2,c:-50,d:8.0}. Asterisks mark spikes.
Acknowledgements
Funding received from the NSF/MRC NeuroNex Odor2Action programme 274 (NSF #2014217, MRC #MR/T046759/1).
References
[1] Schmuker, M., Bahr, V., & Huerta, R. (2016). Exploiting plume structure to decode gas source distance using metal-oxide gas sensors. Sensors and Actuators B: Chemical, 235, 636–646
[2] Ackels, T., Erskine, A., Dasgupta, D., et al. (2021). Fast odour dynamics are encoded in the olfactory system and guide behaviour. Nature, 593(7859), 558–563
[3] Izhikevich, E. M. (2003). Simple model of spiking neurons. IEEE Transactions on Neural Networks, 14(6), 1569–1572

[4] Kepecs, A., Wang, X. J., & Lisman, J. (2002). Bursting neurons signal input slope. Journal of Neuroscience, 22(20), 9053–9062

Speakers

Volker Steuber

Rebecca Miko

Marcus Scheunemann

Michael Schmuker

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P200: Modeling the response of cortical cell populations to transcranial magnetic stimulation

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P200 Modeling the response of cortical cell populations to transcranial magnetic stimulation

Aaron Miller1, Konstantin Weise1,2,Thomas R. Knösche*1,3

1Max Planck Institute for Human Cogntitive and Brain Sciences, Leipzig, Germany

2Leipzig University of Applied Sciences, Germany

3Technical University Ilmenau, Germany

*email: knoesche@cbs.mpg.de
Introduction

The response of cortical neurons to TMS depends on the locally induced electric field magnitude and direction as well as on the physiological, biophysical, and microstructural properties of the involved cells. Here, we provide a modeling framework to integrate standard neural population modeling with numerically estimated TMS induced electric fields, mediated by detailed information on cell morphology and physiology. We exemplify this framework for the stimulation of primary motor cortex (M1), giving rise to observable electromyographic recordings in muscles (motor evoked potentials – MEP) as well as fast activity volleys in the EEG (DI-waves).

Methods
The model comprises pairs of pre/postsynaptic neural populations and their generation of short-latency (<10ms) responses upon TMS. We focus on the generation of I-waves by activation of neurons that project to layer 5 (L5) corticospinal neurons. We use realistic compartment models to simulate spatiotemporal spiking dynamics on the axonal arbors of presynaptic neurons. This output is coupled into L5 cells according the morphologies of presynaptic axonal and postsynaptic dendritic trees. The resulting current entering L5 somata defines an average current input to a neural mass model. We explore their sensitivity towards model parameters using a generalized polynomial chaos (gPC) approach.

Results
Fig. 1A-C show the resulting modeling pipeline. The output activity of L5 neurons due to stimulation of upstream L2/3 neurons is presented in Fig. 1D. We observe a strong directional dependency at low and medium intensity, decreasing at higher intensities, which agrees with experimental and modeling results (Souza et al., 2022; Weise et al., 2023). A gPC surrogate of the activity function using 4000 model evaluations with random parameter distributions resulted in a normalized root mean square deviation of 1.9% tested against 1000 independent verification runs.
The average Sobol indices revealed the most influencing parameters and combinations thereof, i.e. E/I balance (42%), stimulation intensity (13%), and a combination of both (14%).

Discussion
The model provides the basis for modeling TMS evoked activity using parsimonious NMM with high biological detail. Previous coupling models were based on coarse approximations and ignored the complex mechanisms of how TMS activates neuronal populations. The model pipeline can also be adapted to other brain stimulation methods such as tDCS. The calculated surrogate models will be provided for download in order to allow efficient calculation of the input currents to L5 PC.

Figure 1. A: Parameters of the TMS induced electric field; B and C: Illustration of the model pipeline - induced e-field acts on terminals of presynaptic axons. Spreading of activity in axonal arbors is captured by the axonal delay kernel. The postsynaptic synapto-dendritic delay kernel accounts for extra position-dependent delay and yields current entering soma; D: Resulting input current to L5 PC over tim
Acknowledgements
The publication was supported by BMBF grant 01GQ2201 (KW, TRK).
References
K. Weise, T. Worbs, B. Kalloch, V.H. Souza, A.T. Jaquier, W. Van Geit, A. Thielscher, T.R. Knösche: Directional Sensitivity of Cortical Neurons Towards TMS Induced Electric Fields. Imaging Neuroscience 1: 1–22 (2023)

V.H. Souza, J.O. Nieminen, S. Tugin, L.M. Koponen, O. Baffa, R.J. Ilmoniemi: TMS with fast and accurate electronic control: Measuring the orientation sensitivity of corticomotor pathways. Brain Stimulation 15(2), 306–315 (2022)

Speakers

Thomas R. Knösche

Konstantin Weise

Aaron Miller

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P201: HoloNeV: Holographic visualization tool for neural network activity

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P201 HoloNeV: Holographic visualization tool for neural network activity

Safeer A. Mirani1, Pirah Menon1, Rosanna Migliore2, Michele Migliore2, Beniamina Mercante2, Paolo Enrico1, Sergio MG Solinas1*

1Department of Biomedical Sciences, University of Sassari, Sassari, SARDINIA ITALY
2Institute of Biophysics, National Research Council, Palermo, Italy

*Email: smgsolinas@uniss.it

Introduction
Recent neuroscience initiatives have generated extensive data on neuroanatomy and brain function, enabling the development of detailed models of brain areas.While tools like NetPyne[1]and PyNN[2]facilitate network design on NEST[3]or NEURON[4], visualization tools are lagging behind. Emerging 3D holographic devices are bridging the digital and physical worlds instancing interactive virtual objects in the real world, i.e. Mixed Reality. Here, we introduce HoloNeV, a high-performance tool for visualizing and interacting with 3D neural network models. We validate the HoloNeV on a neuronal dataset derived from the CA1 region of the mouse hippocampus[5], encompassing the somata of pyramidal cells and interneurons divided into four neuronal layers.

Methods
We developed HoloNeV in Unity 3D (Version 2022), a game development platform designed for dynamic and high-resolution of complex animations to run on the Microsoft Hololens 2 headset using the Mixed Reality Toolkit for MR integration. We leverage on the technique of GPU instancing to enhance visualization performance, enabling the rapid rendering of numerous neuron representations simultaneously, up to 300.000 somata. We designed a custom low level code to control the data management in the GPU, a.k.a. shader, which supports GPU-accelerated rendering with stereo capabilities, transforming neuron positions and activations into an immersive visual experience.Tests were run on a workstation Intel Xeon w9-3495X with 128 GB RAM and GPU Nvidia RTX A5500.

Results
Following successful hardware integration and software development, the system was tested using the brain hippocampus dataset visualized through Microsoft HoloLens 2. The immersive 3D model allows exploration of neuronal organization in the CA1 region. The visualization includes a 3D distribution of neurons, density patterns, and layer-wise organization and is able to replay the neuronal activity from stored spike trains. While using standard state of art Unity tools the performance was 10 FPS, HoloNeV performance testing showed a mean frame rate of 97.8 FPS, ensuring a comfortable user experience in mixed reality.

Discussion

We introduce HoloNeV, a mixed-reality tool for visualizing neural network activity using an holographic headset. This system allows researchers to interact with neural networks in 3D while staying aware of their physical surroundings. Key innovations include stereo rendering optimization and fine hand-tracking for direct manipulation. Researchers can customize visualization parameters such as neuron size and density in real time. Although currently limited to representing the soma without axon and dendrite, future developments will address the full neuronal morphology, the configuration of neuronal parameters, real-time data streaming from HPC facilities, paving the way for new insights in neuroscience research.

Acknowledgements
Project IR00011 EBRAINS-Italy - Mission 4, “Istruzione e Ricerca” - Component 2, “Dalla ricerca all impresa” - Line of investment 3.1 of PNRR, Action 3.1.1 NextGeneration EU (CUP B51E22000150006) awarded to P. E. and S. S., the “FeNeL” project, PNRR M4.C2.1.1 – PRIN 2022 – No. 2022JE5SK2 – CUP G53D23000380006 awarded to S. S., Project “Numeracy in Aging (NiA) CUP J53D23017580001 awarded to P. E.
References
1. Dura-Bernal, S. et al. (2018) https://doi.org/10.1101/461137
2. Davison, A. P. (2008). https://doi.org/10.3389/neuro.11.011.2008
3. Gewaltig, M.-O., & Diesmann, M. (2007). https://doi.org/10.4249/scholarpedia.1430
4. Hines, M. (2009). https://doi.org/10.3389/neuro.11.001.2009

5. Gandolfi Daniela, et al. (2022).https://doi.org/10.1038/s41598-022-18024-y

Speakers

Sergio Solinas

Rosanna Migliore

Researcher, Istituto di Biofisica - CNR

Computational NeuroscienceEBRAINS-Italy Research Infrastructure for Neuroscience https://ebrains-italy.eu/

Michele Migliore

Safeer Mirani

Pirah Memon

Beniamina Mercante

Paolo Enrico

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P202: Implementation of an SNN-based LLM

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P202 Implementation of an SNN-based LLM

Tomohiro. Mitsuhashi*1, Rin. Kuriyama1, Tadashi. Yamazaki1

1Graduate School of Informatics and Engineering, The University of Electro-Communications, Tokyo, Japan

*Email:m2431154@gl.cc.uec.ac.jp

Introduction

Large language models (LLMs) are indispensable in everyday life and business, yet their training and inference demand an enormous amount of electricity. A major contributor to this consumption is the extensive memory access in artificial neural network (ANN) models. A potential solution is to use neuromorphic hardware, which emulates dynamics of spiking neural networks (SNNs)[1]. SpikeGPT has been proposed as an SNN-based LLM[2]. However, not all components in SpikeGPT are implemented by spiking neurons. In this study, we aimed to implement a fully spike-based LLM based on the present SpikeGPT.

Methods
SpikeGPT consists of two blocks: Spiking RWKV and Spiking RFFN (Fig. 1A). These blocks consist of a component that performs analog computation and another component that converts the results into spike sequences by an SNN. We replaced the former component with an SNN by using the method proposed by Stanojevic[3], that uses spike timing information (Time-to-First Spike, TFS) (Fig. 1B), where an analog value is represented by the time that a neuron emits a spike for the first time. Eventually, we developed SNN-based RWKV and SNN-based RFFN (Fig. 1A). Moreover, nonlinear processes, including the calculation of an exponential function, were approximated using multi-layer SNNs, enabling the entire processing to be implemented solely with SNNs.
Results
We completed implementation of an SNN-based LLM, ensuring that both the RWKV and RFFN blocks are SNNs. Our SNN-based LLM should have generated the same sentences with the original SpikeGPT, but it generated completely broken sentences. We performed a quantitative comparison between analog computation values and approximated ones represented by spike timing, and found discrepancies between them. Namely, the nonlinear processes by SNNs did not work well. Then, we reverted SNN-based nonlinear processes with the original analog versions. We were able to obtain readable sentences, although the sentences were still different (Fig. 1C). Notably, we confirmed that each neuron emitted at most one spike during text generation (Fig. 1D).

Discussion
We implemented an SNN-based LLM that generates sentences. Nonetheless, our SNN-based nonlinear processes need to be improved for better approximation. One possible way is to set the temporal resolution of the SNNs much smaller for finer precision of analog values represented by TFS. Meanwhile, each neuron has at most one spike for each propagation, combining our model with neuromorphic hardware could lead to significant energy savings. These advances are expected to address challenges associated with energy-efficient LLMs.

Figure 1. Overview of our SNN-based LLM and sample results. (A) The architecture of SpikeGPT (left) and our model (right). (B) Schematic of the TFS approach, where the temporal difference between the time parameter and the spike time encodes an analog value. (C) A sample sentence generated by our model. (D) Raster plots for the SNN-based RWKV and SNN-based RFFN during token generation.
Acknowledgements
This study was supported by MEXT/JSPS KAKENHI Grant Numbers JP22H05161, JP22H00460.
References
1. Davies, M., et al. (2021). Advancing neuromorphic computing with Loihi: A survey of results and outlook.Proceedings of the IEEE, 109(5), 911–934.https://doi.org/10.1109/JPROC.2021.3067593

2. Zhu, R.-J., et al. (2024). SpikeGPT: Generative pre-trained language model with spiking neural networks.arXiv preprint.https://arxiv.org/abs/2302.13939

3. Stanojevic, A., et al. (2023). An exact mapping from ReLU networks to spiking neural networks.Neural Networks, 168, 74–88.https://doi.org/10.1016/j.neunet.2023.09.011

Speakers

Tomohiro Mitsuhashi

Rin Kuriyama

Tadashi Yamazaki

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P203: Reciprocity and Hierarchical Organization in the Resting-State Brain: Implications for Efficient Connectivity

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P203 Reciprocity and Hierarchical Organization in the Resting-State Brain: Implications for Efficient Connectivity

Guillermo Montaña-Valverde*1,2, Paula García-Royo2, Wolfram Hinzen1,3, Gustavo Deco2,3

¹ Department of Translation and Language Sciences, Pompeu Fabra University, Barcelona, 08018, Spain
² Center for Brain and Cognition, Computational Neuroscience Group, Department of Information and Communication Technologies, Universitat Pompeu Fabra, Barcelona, 08018, Spain
³ Institució Catalana de Recerca i Estudis Avançats, ICREA, Barcelona, 08010, Spain

*Email: guillermo.montana@upf.edu

Introduction

While the brain is traditionally considered to have a strong hierarchical organization, our findings demonstrate that its structure is more flattened, facilitating more efficient information flow across the network [1]. Using a large resting-state and task-based fMRI dataset, we show that reciprocity – the tendency of a network to have bidirectional connections– strongly correlate with hierarchical organization. This suggests that the brain’s densely interconnected architecture flattens hierarchy, facilitating efficient information flow through shorter average path lengths and enhanced small-worldness. These results aligns with the idea that reciprocity enhances information flow via feedback connectivity [2]. This novel framework lays the foundation for our understanding of whole-brain functional dynamics [3].

Methods
We analyzed open-source fMRI data from the Human Connectome Project (HCP), comprising both resting-state and 7 task-based data from 1000 subjects. Generative effective connectivity (GEC) – an extension of the classic effective connectivity [4] –was estimated from whole-brain modeling for each subject in the DK80 parcellation [5], providing a directed weighted network [6,7]. Hierarchy was then determined by computing measures of coherence and trophic levels on the GEC (Fig. 1a). Reciprocity, defined as the fraction of total connection strength that is bidirectionally shared between regions, captures the balance between feedforward and feedback interactions [8] (Fig. 1b).
Results
We found that the brain in resting-state exhibits a high degree of reciprocity (0.93± 0.02, Fig. 1c), which shows a strong negative correlation with hierarchical coherence (corr=-0.97, p<0.001, Fig. 1d). Conversely, by artificially modulating for more asymmetric interactions, the hierarchy becomes more rigid (Fig. 1e). In addition, a more flattened hierarchy was associated with a shorter average path length (corr=0.97, p<0.001, Fig. 1f), higher average clustering coefficient (corr=-0.95, p<0.001, Fig. 1g), and increased small-worldness (corr=-0.99, p<0.001, Fig. 1h). Furthermore, decreased hierarchical coherence was observed during task performance (Fig. 1i).
Discussion
Overall, our results demonstrate that reciprocity plays a crucial role in shaping the brain’s hierarchical organization. The brain’s nature of high reciprocal connections facilitates information flow and integration, potentially optimizing cognitive processing both at rest and during task performance. On the contrary, a stronger hierarchy, reduces flexibility and adaptability, leading to a worsening in brain connectivity. For this reason, future research in this methodology should explore neuropsychiatric disorders, where changes in hierarchical organization of the brain may underlie altered brain processing. Ultimately, exploring whether targeted interventions that modulate reciprocity can restore optimal hierarchical organization and improve cognitive function.

Figure 1. A. The hierarchy was quantified measuring directedness based on trophic levels. B. Simplified representation of reciprocity. C. Reciprocity in the HCP resting-state dataset. D. Coherence and Reciprocity relation in HCP resting-state. E. Hierarchical representations for different reciprocities. F, G and H. Graph measures correlates with coherence. I. Coherence in 7 tasks compared to resting-state.
Acknowledgements
This study is part of the projectI+D+i Generación de ConocimientoPRE2020-095700, funded by MCIN/AEI/10.13039/501100011033.
References
●
DOI: 10.1038/s41583-023-00756-z
●
DOI: 10.1016/j.neuroimage.2020.117479
●
DOI: http://dx.doi.org/10.1038/s44220-024-00298-y
●
DOI: 10.1016/s1053-8119(03)00202-7
●
DOI: 10.1038/s41562-020-01003-6
●
DOI: 10.1016/j.neuron.2014.08.034
●
DOI: 10.1016/j.celrep.2020.108128
● DOI: 10.1038/srep02729

Speakers

Guillermo Montaña-Valverde

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P204: Brain-like networks emerge from distance dependence and preferential attachment

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P204 Brain-like networks emerge from distance dependence and preferential attachment
Aitor Morales-Gregorio*1, Karolína Korvasová1

1Faculty of Mathematics and Physics, Charles University, Prague, Czechia

*Email: aitor.morales-gregorio@matfyz.cuni.cz
Introduction
Neurons in the brain are not randomly connected to each other. Neuronal networks have low density, high local clustering, short path lengths, heavy-tailed weight and degree distributions, and distance-dependent connection probability. These properties enable efficient information processing. However, standard network generating algorithms cannot produce networks to show all these brain-like properties. Here, we show that distance-dependent connection probability in combination with preferential attachment can generate brain-like networks that match the properties of the neuronal networks from six animal species: C. Elegans [1], Platynereis [2], Drosophila [3,4], Mouse [5,6], Marmoset [7], and Macaque [8].

Methods
Networks are created by iterative growth, mimicking how neurons would naturally grow. Neurons are randomly positioned inside a sphere, the distance between them calculated, and multiplied by the exponential kernel [9], thus creating the distance-dependent probability. An empty network is initialized, to which a new connection is added in each iteration drawn from the distance-dependent probability. The iteration stops when the target density is reached.
To achieve heavy-tailed distributions we study preferential attachment, i.e. a higher probability of connection for edges with high weight (weight-preferential) or nodes with high degree (degree-preferential).

Results
The neuronal networks of six animals have low density, high local clustering, short global path lengths, and heavy-tailed weight and degree distributions.
We show that distance dependence alone can create small-world networks with high clustering and short path lengths, but fails to produce heavy-tailed weight or degree distributions. Including weight-preferential attachment enables the creation of networks that also have heavy-tailed weight distributions, but not of the degrees. Finally, we show that degree-preferential attachment together with distance dependence produces brain-like networks that simultaneously have all the mentioned properties, and can match the experimentally measured networks of six different animal species.

Discussion
Our algorithm can match the properties of the neuronal networks of six different animals, suggesting these could be general principles of neural network development. It is well-known that neurons at large distances are less likely to be connected, in part because these connections are metabolically more expensive to establish and maintain than short-range ones. The large neuropil branching of some neurons increases the probability of connections with them, which we capture via the degree-preferential mechanism.
In conclusion, distance dependence and preferential attachment are biologically realistic mechanisms that can produce networks closely matching both invertebrate and vertebrate brains.

Acknowledgements
This work received funding from the Programme Johannes Amos Comenius (OP JAK) under the project 'MSCA Fellowships CZ - UK3' (reg. n. CZ.02.01.01/00/22\_010/0008220); and from Charles University grant PRIMUS/24/MED/007
References
[1] Varshney et al (2011) PLoS CB 7:e1001066
[2] Randel et al (2014) eLife 3:e02730
[3] Takemura et al (2013) Nature 500:175-181
[4] Scheffer et al (2020) eLife 9:e57443
[5] MICrONs Consortium et al (2021) bioRxiv 2021.07.28.454025
[6] Gămănuţ et al (2018) Neuron 97(3):698-715
[7] Majka et al (2020) Nature Communications 11:1133
[8] Markov et al (2014) Cerebral Cortex 24(1):17-36
[9] Ercsey-Ravasz et al (2013) Neuron 80(1):184-197

Speakers

Aitor Morales-Gregorio

Karolína Korvasová

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P205: High-frequency oscillations in primate visual areas: Critical insights into neural population dynamics or mere spike artifacts?

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P205 High-frequency oscillations in primate visual areas: Critical insights into neural population dynamics or mere spike artifacts?
Katarína Studeničová*1, Aitor Morales-Gregorio1, Karolína Korvasová1

1Faculty of Mathematics and Physics, Charles University, Prague, Czech Republic

*Email: katarina.studenicova@matfyz.cuni.cz

Introduction
Short bursts of high-gamma frequency oscillations (80-150 Hz) resembling hippocampal ripples were to date observed in several cortical areas [1-4]. However, their function and link to memory processes are unclear. The goal of our work is to describe the relationship between the high-gamma bursts, referred to as cortical ripples, and neuronal spiking activity in the same cortical location under different levels of drowsiness.
Methods
We analyze a dataset of the resting state activity consisting of 4 macaque monkeys ([5,6], and additional data provided by the authors), each having 16 Utah arrays implanted in visual areas V1, V2, V4, and IT. Monkeys sat in a dark room, and the vigilance of the recorded animals changed, ranging from fully alert to drowsy and light sleep. Raw traces were downsampled and filtered to the ripple band signal (80-150 Hz), and spikes were sorted. Short high-amplitude oscillatory bursts present in the ripple band, further referred to as cortical ripples, were detected by the standard double thresholding methods, and additionally confirmed by spectral analysis and surrogate methods.

Results
During the alert eyes open states without strong visual input, the dynamic of a network is unorganized in both space and time. However, with increasing drowsiness, the network falls into a global upstate-downstate regime. Upstates are strongly visible mainly in V1 and V2, and less organized in V4 and IT, possibly reflecting a different organizational structure of higher cortical areas. In all brain states, cortical ripples are accompanied by spiking activity. In general, spikes are locked to the phase of the ripple band. Most are locked to the trough, however, we also found cells preferring peaks of the oscillatory signal. We detail these findings further by describing a variety of spiking preferences with respect to the ripple band.

Discussion
To the best of our knowledge, we are the first to uncover the global organization of high-frequency oscillatory activity in the macaque visual areas during resting state, spanning large horizontal distances with intracortical recording precision. We prove the existence of cortical ripples in all the areas covered (previously literature only addressed V1 and V4) and describe the relationship between spikes and cortical ripples with respect to various brain states. We detail our findings by area-wise description, highlighting crucial differences. This work aims to bridge gaps between various recording techniques by providing a detailed view of network states underlying high-frequency oscillatory bursts.

Acknowledgements
This work received funding from the Charles University grant PRIMUS/24/MED/007; and the Programme Johannes Amos Comenius (OP JAK) under the project 'MSCA Fellowships CZ - UK3' (reg. n. CZ.02.01.01/00/22\_010/0008220).
References
[1] https://doi.org/10.1523/JNEUROSCI.0742-22.2022
[2] https://doi.org/10.7554/eLife.68401
[3] https://doi.org/10.1093/brain/awae159
[4] https://doi.org/10.1073/pnas.2210698120
[5] https://doi.org/10.1038/s41597-022-01180-1
[6] https://doi.org/10.1016/j.neuron.2024.12.003

Speakers

Aitor Morales-Gregorio

Karolína Korvasová

Katarína Studeničová

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P206: Synaptic Topography Influences Dendritic Integration in Drosophila Looming Responsive Descending Neurons

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P206 Synaptic Topography Influences Dendritic Integration in Drosophila Looming Responsive Descending Neurons

Anthony Moreno-Sanchez*1, Alexander N. Vasserman1,HyoJongJang2, Bryce W. Hina2, Catherine R. von Reyn1 2, Jessica Ausborn1

1Department of Neurobiology and Anatomy, Drexel University College of Medicine, Philadelphia, United States.
2School of Biomedical Engineering, Science and Health Systems, Drexel University, Philadelphia, PA, United States.

*Email: am4946@drexel.edu
Introduction

Synapse organization plays a crucial role in neural computation, affecting dendritic integration and neuronal output [1, 2].InDrosophila melanogaster, visual projection neurons (VPNs) encode distinct visual features and relay retinotopic information from thelobulaandlobulaplate to descending neurons (DNs) in the central brain[3].DNsintegratespatially organizedvisualinformationfrom VPNsto elicitappropriatemotorresponses[4,5,6].However, the retinotopic organization of VPN-DN connections and its impact on dendritic integration remain unclear.Using electron microscopy (EM)data, computational modeling, and electrophysiology, we investigated how synaptic topography affects dendritic processing in looming-sensitive DNs.

Methods
We analyzed EM reconstructions ofDrosophilaVPN-DN circuits from theFull Adult Fly Brain(FAFB)dataset[7], using flywire.ai[8].Wedeveloped multicompartment models of 5 DNs with precise VPN synaptic locationsusingthe FAFB dataset.Using EM VPNmorphologies,we estimated the receptive fields of 6 VPN populations [4,9] and analyzed synapse organization on DN dendrites.We experimentally determined the spike initiation zone (SIZ), in our DNs of interest by tagging the endogenous voltage gated sodium channelpara.Passive properties of DNsweredeterminedusing whole-cell patch clamp electrophysiology data, by fittinghyperpolarizingexperimentalcurrent injections.Simulations were performed in the NEURONsimulationenvironment.
Results
VPN synapses formed spatially constrained clusters on DN dendritesbut lacked retinotopic organizationwithin the clusters.We found that DN morphologyand passive propertiesfilterexcitatory postsynaptic potentials (EPSPs)to achieve synaptic democracy,normalizingeach EPSPs impactattheSIZ.SimulationssuggestthatVPN synapsesfollow anear random distribution of synapsesavoiding tight clusters of synapses from individual neurons to avoid shunting.This synaptic topography,together with synaptic democracy,maintainsa linear relationship between synapse number and depolarization at the SIZ, both when activating individual VPNsand a small group of VPNs.
Discussion
DNs integrate retinotopic feature information from multiple VPN types, each targeting distinct dendritic regions.This organization strategy may enable DNs toselectively process visual features across the fly’s visual field for behavior-relevant computations.Our resultssuggestthat DN dendritic architecture and synaptic topography supports a quasi-linear integration model, in which synaptic democracy ensures consistent encoding of stimulus location via synapse numbers. These findings offer insights into synaptic organization principles andtheirrole in neural circuit function, highlighting the absence of retinotopic organization to prevent membrane shunting.

Acknowledgements
We thank Arthur Zhao forhelpwith the receptive field mapping, James M. Jeanne for help with the creation of dendrograms, and Thomas A. Ravenscroft for providing us with para-GFSTF tools for SIZ labeling.This study was supported in part by the National Institutes of Health (NINDS R01NS118562 to J.A. andC.R.v.R.), and the National Science Foundation (grant no. IOS-1921065 toC.R.v.R.).
References
1.doi:10.1038/s41583-020-0301-7
2.doi:10.1126/science.1189664
3.doi: 10.7554/eLife.21022
4.doi: 10.1038/s41586-023-05930-y
5.doi: 10.1038/nn.3741
6.doi: 10.1016/j.cub.2008.07.094
7.doi: 10.1016/j.cell.2018.06.019.
8.doi:10.1038/s41592-021-01330-0
9.doi: 10.7554/eLife.57685

Speakers

Anthony Moreno-Sanchez

Alexander Vasserman

HyoJong Jang

Bryce Hina

Catherine von Reyn

Jessica Ausborn

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P207: From evolution to creation of spiking neural networks using graph-based rules

P210: Biochemically detailed modelling of cortical synaptic plasticity: the effects of timing of neuromodulatory inputs on LTP/LTD

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P210 Biochemically detailed modelling of cortical synaptic plasticity: the effects of timing of neuromodulatory inputs on LTP/LTD

Tuomo Mäki-Marttunen*1, Verónica Mäki-Marttunen2

1Faculty of Medicine and Health Technology, Tampere University, Tampere, Finland
2NORMENT, Division of Mental Health and Addiction, Oslo University Hospital, Institute of Clinical Medicine, University of Oslo, Norway

*Email: tuomo.maki-marttunen@tuni.fi
Introduction

Synaptic plasticity is a time-sensitive phenomenon. The timing of action potentials in pre- and postsynaptic neurons is well known to influence plasticity outcomes through Hebbian spike-timing-dependent plasticity mechanisms. It is also known that the neuromodulatory state of the neuron strongly affects plasticity [1]. However, it is not fully understood how the timing of neuromodulatory inputs to the cells affects plasticity outcomes [2]. Neuromodulatory activity is important for learning and memory consolidation [3], and thus, understanding how neuromodulatory inputs interact with neuronal activity will allow to gain a deeper view of how brain plasticity is regulated at a higher level [4-6].
Methods
Here, we use a multi-pathway model of synaptic plasticity in the cortex [7-8] to study the interaction between the timing of neuromodulatory and Ca2+inputs to the postsynaptic spine in shaping synaptic plasticity. We investigate how different forms of plasticity are affected by the exact timing of neuromodulatory inputs from the locus coeruleus, which is the main source of norepinephrine (NE) in the mammalian brain, relative to high-frequency Ca2+inputs.
Results
We show that when Ca2+inputs are followed by NE inputs, strong LTP can be observed, whereas LTD occurs when Ca2+inputs followed NE inputs. This effect is caused by a difference in the amount of cAMP produced and PKA activated between the two stimulation protocols: the Ca2+-> NE protocol induces strong PKA activation and GluR1 exocytosis, while the NE -> Ca2+protocol yields much smaller PKA activation.
Discussion
Animal studies suggest that the timing of fast activation of neuromodulatory centers is important [9] and may play a role in the oscillatory processes that underlie memory consolidation during sleep [10]. In addition, recent studies suggest that neuromodulatory activity at slower time scales during sleep presents a timed relation with oscillatory events underlying memory consolidation [11-12]. Our results suggest that a timely activation of locus coeruleus within a wave of brain activity can be crucial for the plasticity outcome, which can have important implications for our understanding of learning and memory consolidation.

Acknowledgements
Funding: Academy of Finland (330776, 358049). The authors also wish to acknowledge CSC Finland (project 2003397) for computational resources.
References

[1] https://doi.org/10.1016/j.neuron.2007.08.013
[2]https://doi.org/10.1038/s41583-020-0360-9
[3]https://doi.org/10.1016/j.neuron.2023.03.005
[4] https://doi.org/10.3389/fnsyn.2016.00038
[5] https://doi.org/10.3389/fncom.2018.00049
[6] https://doi.org/10.3389/fncir.2018.00053
[7] https://doi.org/10.7554/eLife.55714
[8]https://doi.org/10.1073/pnas.231251112
[9] https://doi.org/10.1016/j.conb.2015.07.004
[10] https://doi.org/10.1093/cercor/bhr121
[11] https://doi.org/10.1038/s41593-022-01102-9
[12] https://doi.org/10.1016/j.cub.2021.09.041

Speakers

Verónica Mäki-Marttunen

Tuomo Mäki-Marttunen

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P211: Modeling Burst Variability in Beta-Gamma Oscillations Through Layer-Speciﬁc Inhibition

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P211 Modeling Burst Variability in Beta-Gamma Oscillations Through Layer-Speciﬁc Inhibition

Manoj Kumar Nandi*1,2 , Farzin Tahvili1,2, Clément Gossi-Denjean1,2, Emmanuel Procyk1,2, Charlie Wilson1,2, Matteo di Volo1,2

1Université Claude Bernard Lyon 1, Lyon, Rhône-Alpes, France
2INSERM U1208 Institut Cellule Souche et Cerveau, Bron, France

*Email: manoj.phy09@gmail.com, manoj-kumar.nandi@univ-lyon1.fr

Introduction:Cognitive functions rely on collective neuronal oscillations, captured by EEG/LFP. Beta (13-30 Hz) and gamma (30-100 Hz) oscillations are linked to cognition [1]. These oscillations occur in short bursts with variable frequencies, challenging trial averages and simpliﬁed models. In Ref. [2] we showed spiking and neural mass models reproduce gamma bursts but not variability. In a recent study using the adaptive exponential integrate-and-ﬁre (AdEx) model with diﬀerent percentages of somatostatin (SOM) and parvalbumin (PV), we have shown that SOM/PV density aﬀects oscillation frequencies [3]. Experimental study shows, PV/SOM variability exists across layers [4]. Using AdEx, we model layer-speciﬁc inhibitory variability to explain burst frequency/power variability.

Methods:We analyze experimental LFP data using time-frequency spectrogram analysis to identify bursts, deﬁned as oscillatory events lasting at least two cycles with power exceeding six times the median at that frequency. We extract burst features, including peak frequency, peak power, mean power, duration, frequency span, time of peak power, and burst size. Machine learning methods are applied to assess how these features relate to cognitive processes. We use a computational spiking network model based on the adaptive exponential integrate-and-ﬁre (AdEx) model, incorporating layer-speciﬁc variability in inhibitory populations. This allows us to simulate burst dynamics observed in experimental data and explore how diﬀerent inhibitory neuron densities inﬂuence oscillatory behavior.
Results:From the experimental signal, we ﬁrst calculate the averaged beta and gamma power in the lateral prefrontal cortex (LPFC) across layers. As shown by Ref. [5], we also observed a crossover of powers across layers, where beta power dominates in deep layers, and gamma power dominates in superﬁcial layers. The extracted burst power follows this trend, validating the burst extraction process. Using our model, we replicate this behavior, demonstrating the role of varying inhibitory neuron densities in diﬀerent cortical layers.
Discussion:Our model can exhibit burst dynamics across beta to gamma bands as observed in experimental data. Introducing distinct inhibitory populations (SOM, PV) predicts a cortical hierarchy where increased SOM/PV densities lower oscillation frequencies. Layer-wise modeling reveals burst-like features resembling experimental data. These ﬁndings highlight the importance of inhibitory diversity in shaping oscillatory dynamics and suggest that layer-speciﬁc variability plays a key role in modulating neural activity across frequency bands.

Acknowledgements
This work is supported by the French Ministry of Higher Education (Ministére de l’Enseignement Supérieur) and the project LABEX CORTEX (ANR-11-LABX-0042) of Université Claude Bernard Lyon 1 operated by the ANR.
References
[1]https://doi.org/10.1016/j.neuron.2016.02.028
[2]https://doi.org/10.3389/fncom.2024.1422159
[3]https://doi.org/10.1101/2025.02.23.639719
[4]https://doi.org/10.1038/nn.3446
[5]https://doi.org/10.1038/s41593-023-01554-7

Speakers

Manoj Nandi

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P212: Delayed feedback and the precision of a neural oscillator in weakly electric fish

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P212 Delayed feedback and the precision of a neural oscillator in weakly electric fish

Parisa Nazemi*1,2, John Lewis1,2

¹ Department of Biology, University of Ottawa, Ottawa, Canada
² Brain and Mind Research Institute, University of Ottawa, Ottawa, Canada

*Email: pnaze017@uottawa.ca

Introduction

Precision and reliability of neural oscillations are critical for many brain functions. Among all known biological oscillators, the electric organ discharge (EOD) in wave-type electric fish is the most precise, with sub-microsecond variations in cycle periods and a coefficient of variation of CV ~ 10⁻⁴[1]. The timing of the EOD is set by a medullary pacemaker network comprising 150 neurons with weak electrical coupling. How this pacemaker network achieves such high precision is not clear. One hypothesis is that pacemaker activity is regularized by electrical feedback from the EOD itself.
Methods
To investigate this, we use a computational model of a pacemaker neuron [2] with a delayed auto-feedback current stimulus. The stimulus waveform was chosen to mimic the electric field effects of the EOD. We also use a simple pulse stimulus for comparison.
Results
Our results show that feedback either increases or decreases the CV of the period, depending on the phase of delay:some delays led to low CV (regular oscillations) and others resulted in high CV (variable oscillations), corresponded to distinct regions of the phase response curve (PRC), witha clear relationship between CV and PRC slope (Fig. 1). Specifically, phases associated with the lowest CV (φ_L) and highest CV (φ_H) are near the points where the PRC crosses 1 with positive and negative slopes, respectively.We also tested other neural models [3, 4] with different PRCs and found that, as long as the PRC was type II, the results were similar.
Discussion

These findings provide insights into how time-delayed feedback influence the regularity and sensitivity of neural oscillations. The positive slope of the PRC suggests greater stability and promotes regularity for repeated fixed-delay stimulation. This mechanism could explain how the pacemaker network in weakly electric fish maintains exceptional regularity. More broadly, our findings suggest that feedback-driven stabilization may be a general principle for ensuring precise timing in biological oscillators.

Figure 1. Figure 1. Relationship between the coefficient of variation (CV) of periods with the phase response curve (PRC). A: normalized CV of periods. B: PRC. φ_L and φ_H mark the intersections of the PRC with the baseline at 1, where the slopes are positive and negative, respectively.
Acknowledgements
Supported by an NSERC Discovery Grant to Dr. John Lewis
References

1.https://doi.org/10.1073/pnas.95.8.4684
2.https://doi.org/10.1038/s41598-020-73566-3
3. https://doi.org/10.1523/JNEUROSCI.2715-06.2007
4. https://doi.org/10.7551/mitpress/2526.001.0001

Speakers

Parisa Nazemi

John Lewis

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P213: logLIRA: a novel algorithm for intracortical microstimulation artifacts suppression

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P213 logLIRA: a novel algorithm for intracortical microstimulation artifacts suppression

Francesco Negri*1, David J. Guggenmos2, Federico Barban1,3

1Department of Informatics, Bioengineering, Robotics, System Engineering (DIBRIS), University of Genova, Genova, Italy

2Department of Rehabilitation Medicine and the Landon Center on Aging, University of Kansas Medical Center, Kansas City, KS, United States

3IRCSS Ospedale Policlinico San Martino, Genova, Italy

*Email: francesco.negri@edu.unige.it

Introduction
Intracortical microstimulation is a key tool to study neuropathologies and ultimately develop novel therapies [1, 2]. The analysis of short-latency evoked activity is essential to understand cortical reorganization driven by targeted electrical pulses [1, 3]. However, large voltage fluctuations known as stimulation artifacts hinder recording and analysis of neural response [4-6]. Existing rejection methods struggle with high spatially and temporally variable stimulus artifacts or rely on restrictive assumptions (e.g., absence of signal saturation) [5-8]. We propose a novel algorithm using piece-wise linear interpolation of logarithmically distributed points, alongside a framework to generate a semisynthetic dataset for benchmarking.

Methods
Our method, logLIRA, begins with a 1 ms blanking interval, dynamically extended to the end of signal saturation if present. Interpolation points are then sampled logarithmically, ensuring denser sampling where the signal changes rapidly. Piecewise linear interpolation estimates the artifact which is later subtracted. Possibly remaining secondary artifacts are mitigated by clustering the first 2 ms of recovered signals across trials, averaging and subtracting highly time-locked components. Finally, trial discontinuities are adjusted, and the same spike detection is applied to both ground truth and cleaned data for comparison.

Results
We evaluated logLIRA against three stimulus artifact rejection algorithms (dynamic averaging [5], global polynomial fitting [10], and SALPA [4]) using a semisynthetic dataset as ground truth. Root-mean-square error and cross-correlation at zero lag were calculated for varying mean firing and artifact rates. SALPA and logLIRA outperformed their competitors, excelling in both metrics (Fig. 1A). Notably, logLIRA significantly reduced the blanking interval duration (Fig. 1B), enabling better recovery of short-latency evoked responses while controlling secondary artifacts and thus false positives. Though not fully evident in the semisynthetic dataset lacking direct stimulus-spike correlation, this advantage is obvious in real data (Fig. 1C).

Discussion
With this work we introduced a reliable and effective method for the rejection of stimulus artifacts, highlighting the importance of handling secondary artifacts emerging from a reduced blanking interval or poor suppression due to numerous factors, including signal saturation. A trustworthy recovery of short-latency evoked activity is poised to greatly benefit neuroscientific research: logLIRA could improve the estimation of mesoscale effective connectivity by means of SEEC method [10], aiding in the understanding of cortex stimulation-driven functional reorganization [1, 3], and eventually enhancing the effectiveness of neuroprosthetic systems aimed at treating neuropathologies, improving the life quality of millions of patients [1-3, 11].

Figure 1. Performance comparison of stimulus artifact rejection algorithms on both semisynthetic and real data. A. Cross-correlation at zero lag for different values of mean artifact rate. B. Blanking intervals distribution for logLIRA and SALPA in the benchmark dataset. C. Example of recovered short-latency evoked activity from a real signal. The red vertical bars depict the 1 ms blanking interval.
Acknowledgements
Work supported by #NextGenerationEU (NGEU) and funded by the Italian Ministry of University and Research (MUR), National Recovery and Resilience Plan (PNRR), project MNESYS (PE0000006) - (DN. 1553 11.10.2022).
References
1.https://doi.org/10.1073/pnas.1316885110
2.https://doi.org/10.3389/fnins.2024.1363128
3.https://doi.org/10.1038/nature05226
4.https://doi.org/10.1016/S0165-0270(02)00149-8
5.https://doi.org/10.1016/j.jneumeth.2010.06.005
6.https://doi.org/10.1016/j.jneumeth.2024.110169
7.https://doi.org/10.1371/journal.pcbi.1005842
8.https://doi.org/10.1088/1741-2552/aaa365
9.https://doi.org/10.1088/1741-2552/ab7a4f
10.https://doi.org/10.1016/j.jneumeth.2022.109767

11.https://doi.org/10.3390/brainsci12111578

Speakers

Francesco Negri

David J. Guggenmos

Federico Barban

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P214: A Canonical Microcircuit of Predictive Coding Under Efficient Coding Principles

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P214 A Canonical Microcircuit of Predictive Coding Under Efficient Coding Principles

Elnaz Nemati*1, Catherine E. Davey1, Hamish Meffin1,2, Anthony N. Burkitt1,2

1Department of Biomedical Engineering, The University of Melbourne, Victoria, Australia.
2Graeme Clark Institute, The University of Melbourne, Victoria, Australia.

*Email: fnemati@student.unimelb.edu.au

Introduction

Predictive coding describes how the brain integrates sensory inputs with expectations by minimizing expectation errors [1]. Studies show increased neural activity in cortical L2/3 during sensory mismatches [2], offering insights into disorders like autism [3] and perceptual phenomena such as visual illusions [4]. Canonical microcircuit models [5, 6] advance understanding but often overlook spiking dynamics, detailed inhibitory mechanisms, and Dale’s law adherence. They also neglect the distinct roles of cortical layers, especially L4 and L5/6 [7]. The Deneve framework [8] provides another perspective, modeling neurons as decoders where spikes are triggered if the membrane potentials, representing reconstruction errors, exceeds a threshold.

Methods
This study extends Deneve’s predictive coding framework [7] by assigning Gabor receptive fields to layer 4 neurons, creating a V1-like biologically inspired feature extractor. It introduces two-compartment neurons in layer 2/3 for prediction error signaling within a balanced E/I network. Our hierarchical model mirrors canonical circuits, using spiking neurons and simplified inhibitory populations: Parvalbumin (PV) and Somatostatin (SOM) inspired by Hertäg and Clopath [8]. Layer 5/6 contains similar neurons, generating predictions balanced by these populations (Fig.1a,b). Employing spiking neurons with Leaky Integrate-and-Fire dynamics, the model processes whitened images in ON/OFF channels, as in experimentally observed LGN responses.
Results
The model successfully results in L4 neurons displaying balance (Fig.1c), and orientation and phase selectivity (Fig.1d,e), thereby demonstrating biologically realistic V1 feature extraction. Layer 2/3 neurons robustly signal prediction errors across matched (FF=FB), mismatched (FF≠FB), feedforward-only (FF>FB), and feedback-only (FB>FF) conditions. Neuronal responses matched experimental evidence, where matched inputs minimized activity, while mismatched inputs elicited strong prediction-error signaling (Fig.1d). Critically, layer 5/6 neurons effectively integrated prediction errors from layer 2/3, significantly reducing sensory reconstruction errors and validating their predictive coding function.
Discussion
The model proposes that predictive coding effectively described cortical function through specific feedback interactions within canonical cortical circuits. It highlights the essential roles played by distinct neuronal compartments and inhibitory inter-neuron populations, specifically PV and SOM neurons, in modulating the balance. The close alignment of theoretical predictions with experimental observations supports the model's validity and enhances our understanding of cortical dynamics. Additionally, the model provides a robust foundation for future research in perceptual neuroscience, the development of neuromorphic systems, and the exploration of clinical interventions for disorders involving disrupted predictive coding mechanisms.

Figure 1. Fig 1: (a) Predictive coding microcircuit representation. (b) Detailed circuitry within each layer and connectivity. (c) Display of excitatory, inhibitory, and net currents showing balanced currents (d) Orientation Bias Index and (e) Phase Bias Index of Layer 4 excitatory populations. (f) Spike responses in Layer 2/3 to various feedforward and feedback inputs.
Acknowledgements
-
References
1.https://doi.org/10.1038/4580
2.https://doi.org/10.1016/j.neuron.2020.09.024
3.https://doi.org/10.1152/jn.00543.2015
4.https://doi.org/10.1016/j.neunet.2021.08.024
5.https://doi.org/10.1016/j.neuron.2012.10.038
6.https://doi.org/10.1016/j.neuron.2018.10.003
7.https://doi.org/10.1073/pnas.2115699119

8.https://doi.org/10.1038/nn.4243

Speakers

Elnaz Nemati

Catherine Davey

Hamish Meffin

Anthony Burkitt

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P215: Superior Temporal Interference Stimulation Targeting using Surrogate-based Multi-Goal Optimization

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P215 Superior Temporal Interference Stimulation Targeting using Surrogate-based Multi-Goal Optimization

Esra Neufeld*1, Cedric Bujard1, Melanie Steiner1, Fariba Karimi1,2, Niels Kuster1,2
1IT’IS Foundation, Zürich, Switzerland

2Swiss Federal Institute of Technology (ETH Zurich), Zürich, Switzerland

*Email: neufeld@itis.swiss

Introduction

Temporal Interference (TI) stimulation, an innovative formof transcranial electrical stimulation [1], uses multiple kHz currents with frequency offsets in the brain’s physiological range to steerably and selectivelystimulate deep targets. However, the complex and heterogeneous head environment, along with important inter-subject variability, make it challenging toidentify suitable stimulation parameters. An easy-to-use TI Planning (TIP)application was published [2] to facilitate study design and stimulation personalization. However, due to computational limitations, brute-force explorationof the full parameter space was not feasible, requiring users to impose pre-constrains. This oftenleads to suboptimal settings and making the tool lessaccessible for beginners.

Methods
TIP generates detailed head models from T1-weighted MRI data [3], co-registers the ICBM152 atlas [4] for target region identification, assigns DTI-based anisotropic conductivity maps [5], places electrodes according to the 10-10 system, and performs EM simulations to establish a full E-field basis. Surrogate based optimization (SBO) [6] combines an iteratively-refined Gaussian-process (GP) surrogate and a multi-objective genetic algorithm (MOGA) [7] to identify the front of Pareto-optimal conditions (electrode locations and currents), with regard to the goals of 1) maximizing stimulation strength and 2) selectivity, while 3) avoiding collateral stimulation.

Results
Based on the identified Pareto front, users can interactively weightthe three conflicting goals and compare configurations with comparable perfor-mances based on quantified quality metrics and visualized distributions. Theiterative SBO approachdramatically minimizes the number of full evaluationsrequired to predict the performance metrics (<100 instead of millions), enablingcomprehensive exploration of high-dimensional parameter spaces(5n − 1, n ≥ 2: number of channels).
Discussion
A fully automatic, online accessible tool for personalized TI stimulation planning has been established that leverages AI and image-based simulations. By introducing hybridized, iterative surrogate modeling and MOGA, systematic, comprehensive, and computationally tractable optimization in high-dimensional parameter spaces is achieved and interactive weighting of conflicting objectives becomes possible. The comprehensive search reduces the level of required user expertise, removes arbitrariness, and ensures identification of optimal conditions. The method readily generalizes to non-classic forms of multi-channel TI.

Acknowledgements
---
References
[1]https://doi.org/10.1016/j.cell.2017.05.024
[2] https://tip.itis.swiss
[3]https://doi.org/10.1088/1741-2552/adb88f
[4]https://doi.org/10.1098/rstb.2001.0915
[5]https://doi.org/10.1073/pnas.171473898
[6]https://doi.org/10.1007/978-3-642-20859-1_3
[7] https://doi.org/10.1007/978-981-19-8851-6_31-1

Speakers

Esra Neufeld

Fariba Karimi

Melanie Steiner

Niels Kuster

Cedric Bujard

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P216: Modeling spreading depolarization in control and ischemic neocortical microcircuits using immunostained identify capillaries

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P216 Modeling spreading depolarization in control and ischemic neocortical microcircuits using immunostained identify capillaries

Adam JH Newton*1,Craig Kelley2,Siyan Guo3, Joy Wang3, Sydney Zink4, Marcello DiStasio4,5, Robert A McDougal3,5,6,7, William W Lytton1,8,9,10
●
Department of Physiology and Pharmacology, SUNY Downstate Health Sciences University, Brooklyn, New York
●
Department of Biomedical Engineering, Columbia University, New York, NY.
●
Department of Biostatistics, Yale University, New Haven, CT, United States
●
Department of Pathology, Yale School of Medicine, New Haven, CT, United States
●
Wu Tsai Institute, Yale University, New Haven, CT, United States
●
Department of Biomedical Informatics and Data Science, Yale University, New Haven, CT, United States
●
Program in Computational Biology and Biomedical Informatics, Yale University, New Haven, CT, United States
●
Department of Neurology, SUNY Downstate Health Sciences University, Brooklyn, New York
●
Department of Neurology, Kings County Hospital Center, Brooklyn, New York
●
The Robert F. Furchgott Center for Neural and Behavioral Science, Brooklyn, New York
*Email: adam.newton@neurosim.downstate.edu

Introduction
Brain tissue requires a lot of energy to support the energy-intensive activity of neural information processing, particularly restoring ion homeostasis following action potentials. This high demand for energy leaves the system vulnerable to failures in homeostasis, such as spreading depolarization (SD). SDs are a wave of prolonged depolarizations preceded by a brief period of hyperexcitability that propagates through grey matter at 1-9mm/min [1]. Multiple neurological disorders can lead to SD, including migraine aura, epilepsy, traumatic brain injury, and ischemic stroke.

Methods
We modeled point neurons with Hodgkin-Huxley style channels augmented with homeostatic mechanisms, including Na+/K+-ATPase, NKCC1, KCC2, and dynamic volume changes. Astrocytic buffering was modeled as a field of oxygen-dependent and independent clearance of extracellular K+. Connectivity was based on prior models and the weights and the distribution of external drive were scaled to account for differences between conductance-based Integrate-and-Fire models[2,3]. A 2.0 x 2.3 cm cross-section of the human cortical plate in V1 with immunostaining for CD34, was used to determine the locations of 918 capillaries (mean capillary density: 199.6/cm2; mean±SD capillary cross-sectional area: 16.7±11.9μm2).

Results
We used NEURON/RxD/NetPyNE to simulate13,000 neurons representing ~1 mm3of mouse cortex (layers 2-6), monitoring the concentration of Na+, K+, Cl-, and oxygen, both intra- and extra-cellularly[4–7]. Spreading depolarization could be reliably triggered in each layer by elevating extracellular K+with differences in propagation speed between layers.

Discussion
We use this model to explore the hypotheses that vascular heterogeneity will lead to areas where neurons and astrocytes are well-supplied with oxygen and can better maintain normal activity following insult (increased extracellular K+or reduced perfusion). We also examined the mechanisms that could give rise to greater susceptibility and propagation speeds in the superficial layers compared with the deep cortical layers[8].

Acknowledgements
This research was funded by the National Institute of Mental Health, National Institutes of Health, grant number R01 MH086638, with HPC time fromNIH S10 award, 1S10OD032417-01, and the Yale Center for Research Computing McClearly cluster.

The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
References
1.https://doi.org/10.1007/s12028-021-01429-4
2.https://doi.org/10.1093/cercor/bhs3586
3.https://doi.org/10.1162/neco_a_01400
4.https://doi.org/10.3389/fninf.2022.884046
5.https://doi.org/10.3389/fninf.2018.00041
6.https://doi.org/10.7554/eLife.44494
7.https://doi.org/10.1523/ENEURO.0082-22.2022
8.https://doi.org/10.1177/0271678X16659496

Speakers

Robert McDougal

William W Lytton

Adam J. H. Newton

Craig Kelley

Amy Guo

Joy Wang

Sydney Zink

Marcello DiStasio

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P217: The Generalized Activating Function: Accelerating Axonal Dynamics Modeling for Spinal Cord Stimulation Optimization

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P217 The Generalized Activating Function: Accelerating Axonal Dynamics Modeling for Spinal Cord Stimulation Optimization

Javier García Ordóñez*1,2, Taylor Newton1, Abdallah Alashqar3,4, Andreas Rowald3,4, Esra Neufeld1, & Niels Kuster1,5

1 IT’IS Foundation, Zürich, Switzerland
2 Zürich MedTech AG, Zürich, Switzerland
3 Department of Medical Informatics, Biometry and Epidemiology, Friedrich-Alexander-Universität Erlangen-Nürenberg, Erlangen, Germany
4 Department of Artificial Intelligence in Biomedical Engineering, Friedrich-Alexander-Universität Erlangen-Nürenberg, Erlangen, Germany
5 Swiss Federal Institute of Technology (ETH Zurich), Zürich, Switzerland

*Email: ordonez@itis.swiss

Introduction

The classical Activating Function (AF) provides a fast, linear estimator for membrane polarization as a predictor for stimulation by extracellular electric potential exposure [1]. While computationally efficient, the classical AF fails to account for membrane leakage currents, diffusive interactions between adjacent axonal segments, complex fiber models (multi-cable with periaxonal and paranodal compartments), and the influence of stimulation waveform, limiting its accuracy and usefulness in complex neurostimulation scenarios.

Methods
The Generalized Activating Function (GAF) is a biophysics-based predictor that overcomes these limitations while preserving computational efficiency. The GAF extends the classical framework by convolving the extracellular potential with a Green's function kernel to account for the dynamics of membrane polarization, including axial currents and membrane leakage. A fast Fourier transform is used for the convolutions, producing spike predictions more than 1000× faster than conventional compartmental modeling. The GAF’s formulation accurately predicts dynamic responses in complex fiber models, such as the McIntyre-Richardson-Grill myelinated fiber model [2].
Results
We first verified the GAF by reproducing benchmark experimental and computational data (e.g., strength–duration curves and diameter–dependent rheobase values for different fiber types). Next, we applied the GAF to a clinically validated, realistic model of spinal cord stimulation (SCS)[3]. The GAF’s spike predictions matched those of full electrophysiological simulations, with compute times reduced from hours to seconds. Finally, we leveraged the GAF’s speed and efficiency to explore the design of superior stimulation waveforms and electrode configurations that enhance the selectivity and energy efficiency of SCS. GAF-guided pulse shape optimization discovered charge-balanced waveforms that double recruitment efficacy or reduced power consumption five-fold relative to commonly applied stimulation waveforms.
Discussion
These results demonstrate that the GAF dramatically accelerates neurostimulation modeling without significant loss of accuracy, thereby facilitating large-scale explorations of stimulation parameters and the identification of personalized neuromodulation strategies. By bridging the gap between computational modeling and clinical practice, the GAF paves the way for optimized, patient-specific neurostimulation therapies.

Acknowledgements
No acknowledgements.
References
● https://doi.org/10.1152/jn.00353.2001
● https://doi.org/10.1109/TBME.1986.325670

● https://doi.org/10.1038/s41591-021-01663-5

Speakers

Esra Neufeld

Taylor Newton

Niels Kuster

Abdallah Alashqar

Andreas Rowald

Javier Garcia Ordóñez

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P218: Using AI Technologies To Boost the Simulation and Personalization of Brain Network Models

P222: Computational framework for analyzing oscillatory patterns in neural circuit models

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P222 Computational framework for analyzing oscillatory patterns in neural circuit models

Nikita Novikov1*, Chelsea Ekwughalu1,2, Samuel Neymotin1,3, Salvador Dura-Bernal1,4

1Center for Biomedical Imaging & Neuromodulation, The Nathan Kline Institute for Psychiatric Research, Orangeburg, NY, USA
2Department of Physics, Barnard College, New York, NY, USA
3Department of Psychiatry, NYU Grossman School of Medicine, New York, NY, USA
4Department of Physiology and Pharmacology, State University of New York (SUNY) Downstate Health Sciences University, Brooklyn, NY, USA

*nikknovikov@gmail.com
Introduction

Neural oscillations coordinate brain activity, with abnormal patterns linked to neurological disorders. Understanding their emergence from biological parameters is crucial for effective intervention. While biophysically detailed models provide mechanistic insight, their complexity makes direct analysis computationally expensive and mathematically intractable. To address this, we present a computational framework for the systematic exploration of key parameters governing oscillatory dynamics in large-scale neural networks.
Methods
Our approach relies on the eigenmode decomposition of frequency-dependent transfer matrices, originally proposed in [1] for LIF neurons. In contrast to [1], we do not derive transfer matrices analytically, but instead, we developed a toolbox for their numerical estimation, extending the method to arbitrary models. The estimation is done by automatic construction and simulation of surrogate models, where a single population remains intact, others are replaced by equivalent spike generators, and a sinusoidal signal is added to the probed input. The toolbox is built on top of the NetPyNE framework [2] and supports high-performance parallel simulations.
Results
We validated our approach on a simplified model of cortical layers 2/3 and 4, demonstrating that it accurately decomposes network activity into oscillatory modes and predicts the amplitudes and phases of the oscillations (Fig. 1A-C). Using the computed transfer matrices, we estimated the effects of synaptic weight perturbations by modifying the relevant transfer coefficients and analyzing the resulting eigenmodes, without needing full-model simulations. These predictions closely matched direct simulations of the perturbed model (Fig. 1D, E), confirming that our method reliably identifies key connections that shape oscillatory activity.
Discussion
We propose a framework for systematically exploring the relationship between biological parameters and emergent oscillations. Our tool estimates inter-population transfer coefficients through multiple independent simulations of simple surrogate models, a process well-suited for efficient parallelization. Once computed, these coefficients provide insight into the full model’s oscillatory modes and their sensitivity to parameter perturbations. Our results validate the approach, demonstrating its potential for analyzing neural circuits and informing future neurostimulation and pharmacological interventions.

Figure 1. Figure 1. A – power spectral densities (PSDs. B – eigenmode amplitudes. C – complex relations between L2e and other populations at 60 Hz; arrows – projections of the 1st mode onto populations; black – distribution of simulated instantaneous relations. D, E – effects of L2e->L2i weight perturbation on the 1st mode amplitude (D) and L2i PSD (E).
Acknowledgements
The work is supported by the grants: R01 MH134118-01, RF1NS133972-01, R01DC012947-06A1, R01DC019979, ARL Cooperative Agreement W911NF-22-2-0139, P50 MH109429
References
1. Bos, H., Diesmann, M., & Helias, M. (2016). Identifying Anatomical Origins of Coexisting Oscillations in the Cortical Microcircuit. PLOS Computational Biology, 12(10), e1005132. https://doi.org/10.1371/journal.pcbi.1005132
2. Dura-Bernal, S., Suter, B. A., Gleeson, P., Cantarelli, M., Quintana, A., Rodriguez, F., Kedziora, D. J., Chadderdon, G. L., Kerr, C. C., Neymotin, S. A., McDougal, R. A., Hines, M., Shepherd, G. M., & Lytton, W. W. (2019). NetPyNE, a tool for data-driven multiscale modeling of brain circuits. eLife, 8, e44494. https://doi.org/10.7554/eLife.44494

Speakers

Samuel A. Neymotin

Nikita Novikov

Salvador Dura-Bernal

Chelsea Ekwughalu

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

16:20 CEST

P223: Auto-adjoint method for Spiking Neural Networks

Monday July 7, 2025 16:20 - 18:20 CEST

Passi Perduti

P223 Auto-adjoint method for Spiking Neural Networks

Thomas Nowotny*1, James C. Knight1

1Department of Neuroscience, Example University, City, Country

*Email: t.nowotny@sussex.ac.uk

Introduction
It is important for the success of neuromorphic computing and computational neuroscience to be able to efficiently train spiking neural networks (SNNs). In 2021, Wunderlich and Pehle published the Eventprop algorithm [1], which is based on the adjoint method for hybrid continuous- discrete systems [2]. Eventprop casts the backward pass, which calculates the gradient of a loss function over an SNN, into a hybrid continuous- discrete system of the same nature as the forward dynamics of the SNN. Therefore, Eventprop can be implemented efficiently on both existing SNN software simulators [3] and digital neuromorphic hardware [4].

Methods

Here, we present new work in which we take Eventprop to the next level. The original Eventprop algorithm [1] was derived explicitly for the case of leaky integrate-and-fire (LIF) neurons and “exponential” synapses. The adjoint method for hybrid systems is much more general [2], and [5] already presents a more general set of equations. Here, we choose a level of generality that allows us to derive the general adjoint equations in a form that is explicit enough that the sympy symbolic math Python package can automatically generate code to simulate the equations.

Results
We assume that the neurons of the SNN being trained have internal dynamics described by ordinary differential equations and that their spiking condition and their reset behaviour are described by functions of the neurons’ variables. Finally, we assume that the action caused by an incoming spike entails adding to a neuron variable. Under these general assumptions, we derived a backward pass for adjoint variables like in the original Eventprop and implemented it into our mlGeNN spike-based machine learning framework [6] using sympy. We observe that for leaky integrate-and-fire neurons and exponential synapses, the new framework has the same performance on popular benchmarks as the previous version of standard Eventprop.

Discussion

We have created a new version of mlGeNN that, based on the generalised Eventprop method presented here, allows researchers to rapidly train SNNs with virtually any neuron dynamics using gradient descent with exact gradients. This includes more complex dynamics, such as Hodgkin-Huxley conductance-based models, opening new avenues for injecting function into computational neuroscience models. This new capability is akin to the auto-diff functionality of PyTorch, which has been instrumental in the recent AI revolution.

Acknowledgements
This work was partially funded by the EPSRC, grants EP/V052241/1 and EP/S030964/1.
References
[1] Wunderlich, T. C., & Pehle, C. (2021). Scientific Reports, 11(1), 12829.
[2] Galán, S., Feehery, W. F., & Barton, P. I. (1999). Appl. Num. Math., 31(1), 17-47.
[3] Nowotny, T., Turner, J. P., & Knight, J. C. (2025). Neurom. Comput. Eng., 5(1), 014001.
[4] Gabriel, B., Timo, W., Mahmoud, A., Bernhard, V., Christian, M., & Hector, A. G. (2024). arXiv preprint arXiv:2412.15021.
[5] Pehle, C. G. (2021). Adjoint equations of spiking neural networks (Doctoral dissertation).
[6] Turner, J. P., Knight, J. C., Subramanian, A., & Nowotny, T. (2022). Neurom. Comput. Eng., 2(2), 024002.

Speakers

Thomas Nowotny

James Knight

Monday July 7, 2025 16:20 - 18:20 CEST
Passi Perduti

Posters, Poster session 2

19:10 CEST

Banquet dinner

Monday July 7, 2025 19:10 - 21:40 CEST

TBA

Monday July 7, 2025 19:10 - 21:40 CEST
TBA

Reception/party

08:30 CEST

Registration

Tuesday July 8, 2025 08:30 - 19:00 CEST

Registration/organization

09:00 CEST

Workshops

Tuesday July 8, 2025 09:00 - 10:30 CEST

Rooms 4, 5, 6, 9, 101, 103, 202, 203, Hall 1A, Hall 2B, Hall 3B, Belvedere, Onice, Auditorium, or TBA by workshop organizers

Tuesday July 8, 2025 09:00 - 10:30 CEST
Rooms 4, 5, 6, 9, 101, 103, 202, 203, Hall 1A, Hall 2B, Hall 3B, Belvedere, Onice, Auditorium, or TBA by workshop organizers

Workshops

09:00 CEST

Brain-Inspired Computing

Tuesday July 8, 2025 09:00 - 12:30 CEST

Room 5

Brain-inspired computing looks to mimic how the human brain works to improve artificial intelligence (AI) systems. This area has gained a lot of interest recently because it helps us create stronger and more efficient AI models while tackling challenges faced by current artificial neural networks.

This workshop will cover a range of topics, including biological neural networks, cognitive computing, and biologically-inspired algorithms. We will discuss how learning from the brain's structure and operations can lead to new solutions for complex issues in AI, machine learning, and data processing.

The workshop will include talks from experts in the field and interactive panel discussions. Participants will have the chance to collaborate, share ideas, and connect with others who are excited about using biological principles to advance technology.

Full program in this link.

Schedule
9:00 AM - 9:30 AM Speaker: Rui Ponte Costa, University of Oxford
A theory of self-supervised learning in cortical layers
9:30 AM - 10:00 AM Speaker: Guillaume Bellec, Vienna University of Technology
Validating biological mechanisms in deep brain models with optogenetic perturbation testing
10:00 AM - 10:30 AM Speaker: Guozhang Chen, Peking University
Characteristic differences between computationally relevant features of cortical microcircuits and artificial neural networks
10:30 AM - 11:00 AM Coffee Break
11:00 AM - 11:30 AM Speaker: Robert Legenstein, Graz University of Technology
Rapid learning with phase-change memory-based neuromorphic hardware through learning-to-learn
11:30 AM - 12:00 PM Speaker: Shogo Ohmae, Chinese Institute for Brain Researc
World-model-based versatile computations in the neocortex and the cerebellum
12:00 PM - 12:30 PM Speaker: Yuliang Zang, Tianjin University
Biological strategies for efficient learning in cerebellum-like circuits
12:30 End of Workshop and Lunch Break

Full program in this link.

Speakers

Yunliang Zang

Guozhang Chen

Peking University

Tuesday July 8, 2025 09:00 - 12:30 CEST
Room 5

Workshop

09:00 CEST

Enabling synaptic plasticity, structural plasticity, and mutil-scale modeling with morphologically detailed neurons using Arbor

Tuesday July 8, 2025 09:00 - 12:30 CEST

Belvedere room

Current computational neuroscience studies are often limited to a single scale or simulator, with many still relying on standalone simulation code due to computational power and technology constraints. Simulations incorporating biophysical properties and neural morphology typically focus on single neurons or small networks, while large-scale neural network simulations often resort to point neurons as a compromise to incorporate plasticity and cell diversity. Whole-brain simulations, on the other hand, frequently sacrifice details at the individual neuron and network composition levels.
This workshop introduces recent advances leveraging the next-generation simulator Arbor, designed to overcome these challenges. Arbor enables seamless conversion from the widely used NEURON simulator, facilitates the study of functional and structural plasticity in large neural networks with detailed morphology, and supports multi-scale modeling through co-simulation, integrating microscopic and macroscopic levels of simulation.
Arbor is a library optimized for efficient, scalable neural simulations by utilizing both GPU and CPU resources. It supports the simulation of both individual neurons and large-scale networks while maintaining detailed biophysical properties and morphological complexity. The workshop will feature presentations covering key aspects:

Effortless Transition from NEURON to Arbor - Dr. Beatriz Herrera - Allen Brain Institute, USA
Introducing to the SONATA format, which simplifies the migration process and enables cross-simulator validation, ensuring a smooth transition to Arbor for researchers familiar with NEURON.

Structural Plasticity Simulations - Marvin Kaster & Prof. Felix Wolf - TU Darmstadt, Germany
Presenting ReLEARN and Arbor’s capabilities in modeling distance-dependent structural plasticity, providing insights into structural changes.

Synaptic Plasticity - Dr. Jannik Luboeinski - University of Göttingen, Germany
Showcasing Arbor’s capabilities in modeling calcium-based functional plasticity.

Multi-Scale Co-Simulation with TVB - Prof. Thanos Manos - CY Cergy-Paris University, France
Demonstrating Arbor’s co-simulation with The Virtual Brain (TVB) platform, illustrating the study of epilepsy propagation as an example of multi-scale modeling.

The workshop will conclude with an interactive coding session, offering participants hands-on experience with Arbor and an opportunity to apply the presented concepts.

Speakers

Thorsten Hater

FZJ

Han Lu

postdoc, Forschungszentrum Jülich

Tuesday July 8, 2025 09:00 - 12:30 CEST
Belvedere room

Workshop

09:00 CEST

Inference Methods for Neuronal Models: from Network Activity to Cognition

Tuesday July 8, 2025 09:00 - 12:30 CEST

Room 103

The development of models for neuronal systems have matured in recent years and they exhibit increasing complexity thanks to computer resources for simulation. In parallel, the increasing availability of data poses the challenge to quantitatively related those models to data, going beyond reproducing qualitative activity patterns and behavior. Model inference is thus becoming an indispensable tool for unraveling the mechanisms underlying brain dynamics, behavior, and (dys)function. A critical aspect of this endeavor is the ability to infer changes across multiple scales, from neurotransmitters and synaptic interactions to neural circuits and whole-brain networks. Recent approaches that have been adopted by the neuroscience community include methods for directed effective connectivity (e.g. dynamical causal modeling), simulation-based inference on whole-brain models, and active inference for understanding perception, action and behavior. They have significantly enhanced our ability to interpret data by modeling underlying mechanisms and neuronal processes. This workshop will bring together experts from diverse fields to explore the state-of-the-art methodologies, taking specific applications as examples to compare them and highlight remaining challenges.

Speakers

Matthieu Gilson

chair of junior professor, Aix-Marseille University

Meysam Hahsemi

Research Fellow

Tuesday July 8, 2025 09:00 - 12:30 CEST
Room 103

Workshop

09:00 CEST

Mechanisms for Oscillatory Neural Synchrony

Tuesday July 8, 2025 09:00 - 12:30 CEST

Room 6

https://www.medschool.lsuhsc.edu/cell_biology/cns_2025.aspx

CNS*2025 in Florence, Italy on July 08, 2025From 9:00 to 12:30This workshop will bring together researchers who have recently published on synchronization networks of coupled oscillators, with a mix of approaches but an emphasis on phase response curve (PRC) theory. The researchers come from both theoretical and experimental backgrounds. Topics include synchronization mechanisms for theta nested gamma in the medial entorhinal cortex, mean-field pulsatile coupling methods for fast oscillations in inhibitory networks, beta oscillations in parkinsonian basal ganglia, the relative contributions of synaptic and ultra-fast non-synaptic ephaptic coupling to the inhibition of cerebellar Purkinje cells by basket cells, infinitesimal macroscopic PRC (imPRC) within the exact mean-field theory applied to ING and PING, and robustness in a neuromechanical model of motor pattern generation.
Carmen Canavier, LSU Health Sciences Center New Orleans: “A Mean Field Theory for Pulse-Coupled Oscillators based on the Spike Time Response Curve”
Joshua A Goldberg, Hebrew University of Jerusalem: “Empirical study of dendritic integration and entrainment of basal ganglia pacemakers using phase response curves”
Dimitri M Kullmann, University College London: “Basket to Purkinje Cell Inhibitory Ephaptic Coupling Is Abolished in Episodic Ataxia Type 1”
Hermann Rieke, Northwestern University: “Paradoxical phase response of gamma rhythms facilitates their entrainment in heterogeneous networks”
Yangyang Wang, Brandeis University: “Variational and phase response analysis for limit cycles with hard boundaries, with applications to neuromechanical control problems”
Brandon Williams, Boston University: “Fast spiking interneurons generate high frequency gamma oscillations in the medial entorhinal cortex”

Speakers

Carmen Canavier

Mullins Professor and Department Head, LSU Health Sciences Center NO

Workshop on Mechanisms for Oscillatory Neural SynchronyCNS*2025 in Florence, Italy on July 09, 2025From 14:00 to 17:30This workshop will bring together researchers who have recently published on synchronization networks of coupled oscillators, with a mix of approaches but an emphasis... Read More →

Tuesday July 8, 2025 09:00 - 12:30 CEST
Room 6

Workshop

09:00 CEST

Multiscale Modeling of Electromagnetic Field Perturbations on Neural Activity

Tuesday July 8, 2025 09:00 - 12:30 CEST

Hall 3B

Speakers

Claudia Casellato

Alberto Arturo Vergani

Research Fellow, University of Pavia

Tuesday July 8, 2025 09:00 - 12:30 CEST
Hall 3B

Workshop

09:00 CEST

Neuromodulation, sleep-dependent brain dynamics and information processing

Tuesday July 8, 2025 09:00 - 12:30 CEST

Room 9

Speakers

Michal Zochowski

Sara Aton

Tuesday July 8, 2025 09:00 - 12:30 CEST
Room 9

Workshop

09:00 CEST

NeuroXAI: Explainable AI for Understanding Biological Brains

Tuesday July 8, 2025 09:00 - 12:30 CEST

Onice Room

Speakers

Jie Mei

Nina Hubig

Subham Dey

Claudia Plant

Tuesday July 8, 2025 09:00 - 12:30 CEST
Onice Room

Workshop

09:00 CEST

Cross-species modeling of brain structure and dynamics

Tuesday July 8, 2025 09:00 - 13:00 CEST

Auditorium

Speakers

Eli Müller

James Pang

Research Fellow, Monash University

Tuesday July 8, 2025 09:00 - 13:00 CEST
Auditorium

Workshop

09:00 CEST

Advancing Mathematical Methods in Neuroscience Data Analysis

Tuesday July 8, 2025 09:00 - Wednesday July 9, 2025 12:30 CEST

Hall 2B

Brief Description: With the ever increasing amount of data acquired in neuroscience applications there is an essential need to develop computationally effective, robust, and interpretable data processing algorithms. Recent advancements in graph inference, topology, information theory and deep learning have shown promising results in analyzing biological/physiological data, as well as datasets acquired by intelligent agents. Combining elements from different disciplines of information theory, mathematics, and machine learning is paramount for developing the next generation of methods that will facilitate big data analysis under the realm of better understanding brain dynamics, as well as neuroinspired system dynamics in general. The goal of the workshop is to bring researchers working in data science, neuroscience, mathematics, and machine learning together to discuss challenges posed by analyzing multimodal data sets in neuroscience along with potential solutions, exchange ideas and present their latest work in designing and analyzing effective data processing algorithms. This workshop will serve as a great opportunity to discuss innovative future directions for neuroinspired processing of large amounts of data, while considering novel mathematical data models and computationally efficient learning algorithms.

Schedule:

9:00 - 9:40: Kathryn Hess, EPFL, Topological perspectives on the connectome
Abstract: Over the past decade or so, tools from algebraic topology have been shown to be very useful for the analysis and characterization of networks, in particular for exploring the relation of structure to function. I will describe some of these tools and illustrate their utility in neuroscience, primarily in the framework of a collaboration with the Blue Brain Project.

9:45 - 10:25: Moo Kyung Chung, University of Wisconsin, Topological Embedding of Dynamically Changing Brain Networks
Abstract: We introduce a novel topological framework for embedding time-varying brain networks into a low-dimensional space. Our Topological Embedding captures the evolving structure of functional connectivity by mapping dynamic birth and death values of topological features (connected components and cycles) into a 2D plane. Unlike traditional analyses that rely on synchronized time-points or direct comparisons of network matrices, our method aligns the dynamic behavior of brain networks through their underlying topological features, thus offering invariance to temporal misalignments and inter-subject variability. Using resting-state functional magnetic resonance images (rs-fMRI), we demonstrate that the topological embedding reveals stable 0D homological structures and fluctuating 1D cycles across time, which are further analyzed in the frequency domain through the Fourier Transform. The resulting topological spectrograms exhibit strong associations with age and cognitive traits, including fluid intelligence. This study establishes a robust and interpretable topological representation for the analysis of dynamically changing brain networks, with broad applicability in neuroscience and neuroimaging-based biomarker discovery. The talk is based on arXiv:2502.05814

10:30 - 11:00: Coffee Break

11:00 - 11:40: Anna Korzeniewska, Johns Hopkins University, From causal interactions among neural networks to significance in imaging brain tumor metabolism.

Abstract: Neural activity propagates swiftly across brain networks, often not providing enough data-points to model its dynamics. This limitation can be overcome by using multiple realizations, or repetitions, of the same process. However, once repetitions have been consumed for modeling, or only one is available, the significance of the neural dynamics cannot be assessed using traditional statistical methods. We propose a new method for assessing statistical confidence using the variance of a smooth estimator and a criterion for the choice of a smooth ratio. We show their applications to event-related neural propagations among eloquent and epileptogenic networks, and to metabolite kinetics in hyperpolarized 13C MRI (hpMRI) of brain tumor. The event-related causality (ERC) method - a multichannel extension of the Granger causality concept – was applied to multi-channel EEG recordings to estimate the direction, intensity, and spectral content of direct causal interactions among brain networks. A two-dimensional (2D) moving average, with a rectangular smooth window, sliding over points in the time-frequency plane, provided the smooth estimator and its error for statistical testing. The smooth size of the 2D moving average was determined by the W-criterion, which combines the difference between the smooth estimator and the real values with the confidence interval. The same approach was applied to 2D images of hpMRI of pyruvate metabolism of malignant glioma. A newly developed bivariate smoothing model ensured precise embedding of ERC’s statistical significance in time-frequency space, revealing complex frequency-dependent dynamics of causal interactions. The strength and pattern of neural propagations among eloquent networks reflected stimulus modality, lexical status, and syllable position in a sequence, uncovering mechanisms of speech control and modulation. The strength and pattern of high-frequency interactions among epileptogenic networks identified seizure onset zones and unveiled propagations preceding seizure onset. Statistical confidence of the difference between metabolic responses of tumor and normal tissue, obtained through hpMRI, allowed tumor delineation. Moving average provides an efficient smooth estimator and its error (optimal for reducing random noise while retaining sharp step response) and ensures precise embedding of statistical significance in two-dimensional space. The new approach overcomes several limitations of previously used 2D spline interpolation (restraint to a mesh of knots introducing artifactual distributions of variance and significance, and failure to converge in some cases), while W-criterion provides efficient choice of smooth size. The new technique has broad applicability to neuroscientific research and clinical applications, including planning for epilepsy surgery, localizing anatomical targets for responsive neuromodulation, and gauging tumor treatment response.

11:45 - 12:25: Vasileios Maroulas, University of Tennessee Knoxville, The Shape of Uncertainty.

Abstract: How does the brain know where it is and where it is going? Deep within our neural circuits, specialized cells—like head direction and grid cells—fire in intricate patterns to guide spatial awareness and navigation. But decoding these patterns requires tools that can keep up with the brain’s complexity. In this talk, I will share how we wre using topological deep learning to do just that. Our new models tap into higher-dimensional structures to predict direction and position—without relying on hand-crafted similarity measures. But that is just the beginning. I will also introduce a Bayesian framework for learning on graphs using sheaf theory, where uncertainty is not a bug but a feature. By placing probability distributions on the rotation group and learning them through the network, we gain robustness, flexibility, and accuracy—especially when data is scarce. Together, these advances point to a bold new direction: using geometry and topology to unlock the brain’s code and reshape how we learn from complex data.

Speakers

Vasileios Maroulas

Professor of Mathematics, University of Tennessee Knoxville

topological machine learning, Bayesian computational statistics, manifold learning

Dave Boothe

Neuroscientist, Army Research Laboratory

Ioannis Schizas

Research Engineer, Army Research Lab

Piotr Franaszczuk

Brittany Story

Tuesday July 8, 2025 09:00 - Wednesday July 9, 2025 12:30 CEST
Hall 2B

Workshop

09:00 CEST

Linking structure, dynamics, and function in neuronal networks: old challenges and new directions

Tuesday July 8, 2025 09:00 - Wednesday July 9, 2025 12:30 CEST

Room 101

The full program including abstracts can be found here

https://sites.google.com/view/cns2025workshop-strudynfun/

We are looking forward to seeing you at the workshop.

Wilhelm Braun, Kayson Fakhar and Claus C. Hilgetag

Speakers

Wilhelm Braun

Junior Research Group leader, CAU Kiel, Department of Electrical and Information Engineering

Kayson Fakhar

Claus C Hilgetag

Professor, University Medical Center Eppendorf Hamburg, Germany

Tuesday July 8, 2025 09:00 - Wednesday July 9, 2025 12:30 CEST
Room 101

Workshop

09:00 CEST

Computational strategies in epilepsy modelling and seizure control

Tuesday July 8, 2025 09:00 - Wednesday July 9, 2025 17:30 CEST

Room 202

Epilepsy remains a complex neurological condition, necessitating innovative approaches to understanding
and mitigating seizure activity. This workshop is designed to bring together computational neuroscientists
and researchers with experimental and clinical background to explore cutting-edge strategies in epilepsy
modeling and seizure control. For the general content structure, we plan to start from a modeler's
perspective and then progressively move towards more data-driven approaches.

The first session will explore seizure mechanisms through biophysical and neural mass models at different
temporal and spatial scales, investigating, among others, ionic dynamics and network plasticity. It aims to
understand seizure initiation, progression, and duration.

The second session will focus on the application of computational models to EEG data recorded in epileptic
patients. First, it will discuss advanced parameter inference methods to tailor models to individual data
samples to provide mechanistic insight. It then moves on to issues of seizure monitoring using wearable
devices and long-term EEG recordings, and in particular the use of data features inspired by concepts
derived from mathematical modeling in epilepsy.

The third session will examine stimulation-based strategies to terminate or prevent seizures. There will
be a focus on recent advancements in closed-loop and low-frequency electrical stimulation to control
seizures. On top of model-based approaches, this session will also include the clinical perspective on
stimulation treatment and data-driven studies.

Speakers

Helmut Schmidt

Scientific researcher, Institute of Computer Science, Czech Academy of Sciences

Jaroslav Hlinka

Senior researcher, Institute of Computer Science of the Czech Academy of Sciences

Guillaume Girier

Postdoc, INSTITUTE OF COMPUTER SCIENCE The Czech Academy of Sciences

Isa Dallmer-Zerbe

Tuesday July 8, 2025 09:00 - Wednesday July 9, 2025 17:30 CEST
Room 202

Workshop

09:00 CEST

Population activity: the influence of cell-class identity, synaptic dynamics, plasticity and adaptation

Tuesday July 8, 2025 09:00 - Wednesday July 9, 2025 17:30 CEST

Room 4

Title: Population activity : the influence of cell-class identity, synaptic dynamics, plasticity and adaptation.

Organizers:
Michele GIUGLIANO (co-organizer)
Università degli Studi di Modena e Reggio Emilia - Dipartimento di Scienze Biomediche, Metaboliche e Neuroscienze sede ex-Sc. Biomediche - Italy
michele.giugliano@unimore.it

Simona OLMI (co-organizer)
Institute for Complex Systems - National Research Council - Italy
simona.olmi@fi.isc.cnr.it

Alessandro TORCINI (co-organizer)
Laboratoire de Physique Théorique et Modélisation - CY Cergy Paris Université- Cergy-Pontoise, France
alessandro.torcini@cyu.fr

Abstract:
In recent years tremendous developments have been achieved in the comprehension of neural activity at the population level. This has been possible on one side thanks to the new investigation methods recently developed (e.g. the neuropixels probes and large-scale imaging) that allows for the contemporary registration of the activity of (tens/hundreds of) thousands of neurons in alive and behaving mice as well as established dynamic-clamp protocols.
On the other side by the elaboration of extremely refined mean field models able to describe the population activity of spiking neural networks encompassing realistic biological features, from different forms of synaptic dynamics to plastic and adaptive aspects present at the neural level.
The aim of this workshop is to gather neuroscientists, mathematicians, engineers, and physicists all working on the characterization of the population activity from different point of views, ranging from data analysis of experimental results to simulations of large ensembles of neurons, from next generation neural mass models to dynamical mean field theories. This workshop will favour the exchanges and the discussion on extremely recent developments in this extremely fluorishing field.

Key Words : Neuropixels probes; neural mass models; Fokker Planck formulation; dynamical mean field theory; short- term and long-term plasticity; excitatory and inhibitory balanced networks; spike frequency adaptation

Program:
July 8th -- Room 4
9:15-9:30 Opening
9:30-10:00 Anna Levina (University of Tübingen, Germany)
Talk title: "Balancing Excitation and Inhibition in connectivity and synaptic strength"
10:00-10:30 Giacomo Barzon (Padova Neuroscience Center, University of Padova, Italy)
Talk title: "Optimal control of neural activity in circuits with excitatory-inhibitory balance"

10:30-11:00 Coffee break

11:00-11:30 Eleonora Russo (Scuola Superiore Sant'Anna, The BioRobotics Institute, Italy)
Talk title “Integration of rate and phase codes by hippocampal cell-assemblies supports flexible encoding of spatiotemporal context”
11:30-12:00 Tobias Kühn (University of Bern, Switzerland)
Talk title: "Discrete and continuous neuron models united in field theory: statistics, dynamics and computation"
12:00-12:30 Gianluigi Mongillo (Sorbonne Université, INSERM, CNRS, Institut de la Vision, F-75012 Paris, France)
Talk title: “Synaptic encoding of time in working memory”

July 9th -- Room Hall 1A
9:30 - 10:00 Magnus J.E. Richardson (Warwick Mathematics Institute, UK)
Talk title: "Spatiotemporal integration of stochastic synaptic drive within neurons and across networks"
10:00-10:30 Gianni Valerio Vinci (Istituto Superiore di Sanita’, Rome, Italy)
Talk title: "Noise induced phase transition in cortical neural field: the role of finite-size fluctuations"

10:30-11:00 Coffee break

11:00-11:30 Simona Olmi (Institute for Complex Systems - National Research Council - Italy)
Talk title: “Relaxation oscillations in next-generation neural masses with spike-frequency adaptation”
11:30-12:00 Ferdinand Tixidre (CY Cergy Paris University, France)
Talk title: "Is the cortical dynamics ergodic? A numerical study in partially-symmetric networks of spiking neurons"
12:00-12:30 Letizia Allegra Mascaro (Neuroscience Institute, National Research Council, Italy)
Talk title: "State-Dependent Large-Scale Cortical Dynamics in Neurotypical and Autistic Mice"

12:30-14:00 Lunch break

14:00-14:30 Alessandro Torcini (CY Cergy Paris Université- Cergy-Pontoise, France)
Talk title : “Discrete synaptic events induce global oscillations in balanced neural networks"
14:30-15:00 Rainer Engelken (Columbia University, NY, United States)
Talk title:"Sparse Chaos in Cortical Circuits: Linking Single-Neuron Biophysics to Population Dynamics"
15:00-15:30 Tilo Schwalger (Technische Universität Berlin, Institut für Mathematik, Germany)
Talk title: "A low-dimensional neural-mass model for population activities capturing fluctuations, refractoriness and adaptation"

15:30-16:00 Coffee break

16:00-16:30 Giancarlo La Camera (Stony Brook University, NY, United States)
Talk title: “Prefrontal population activity during strategic behavior in context-dependent tasks”
16:30-17:00 Gorka Zamora-López (Universitat Pompeu Fabra, Barcelona, Spain)
Talk title: "Emergence and maintenance of modular hierarchy in neural networks driven by external stimuli"
17:00-17:30 Sacha van Albada (Research Center Juelich and University of Cologne, Germany)
Talk title: "Determinants of population activity in full-density spiking models of cerebral cortex"

Speakers

Michele GIUGLIANO

Simona OLMI

Alessandro TORCINI

Professor, CY Cergy Paris Universite'

Theoretical physicist interested in #complexsystems #computationalneuroscience #Neuroscience #statisticalphysics #synchronization #neuroskyence #compneurosky #complexity@torcini.bsky.social@torcini@mastodon.social... Read More →

Tuesday July 8, 2025 09:00 - Wednesday July 9, 2025 17:30 CEST
Room 4

Workshop

09:00 CEST

Workshop on Methods of Information Theory in Computational Neuroscience

Tuesday July 8, 2025 09:00 - Wednesday July 9, 2025 17:30 CEST

Room 203

Workshop website: https://kgatica.github.io/CNS2025-InfoTeory-W.io/

Methods originally developed in Information Theory have found wide applicability in computational neuroscience. Beyond these original methods there is a need to develop novel tools and approaches that are driven by problems arising in neuroscience. A number of researchers in computational/systems neuroscience and in information/communication theory are investigating problems of information representation and processing. While the goals are often the same, these researchers bring different perspectives and points of view to a common set of neuroscience problems. Often they participate in different fora and their interaction is limited. The goal of the workshop is to bring some of these researchers together to discuss challenges posed by neuroscience and to exchange ideas and present their latest work. The workshop is targeted towards computational and systems neuroscientists with interest in methods of information theory as well as information/communication theorists with interest in neuroscience.

This is the 20th iteration of this workshop at CNS --join us to celebrate!

Speakers

Joseph T. Lizier

Associate Professor, Centre for Complex Systems, The University of Sydney

My research focusses on studying the dynamics of information processing in biological and bio-inspired complex systems and networks, using tools from information theory such as transfer entropy to reveal when and where in a complex system information is being stored, transferred and... Read More →

Abdullah Makkeh

Postdoc, University of Goettingen

My research is mainly driven by the aim of enhancing the capability of information theory in studying complex systems. Currently, I'm focusing on introducing novel approaches to recently established areas of information theory such as partial information decomposition (PID). My work... Read More →

Pedro A.M. Mediano

Marilyn Gatica

Postdoctoral Research Assistant, Northeastern University London

Michael Wibral

Tuesday July 8, 2025 09:00 - Wednesday July 9, 2025 17:30 CEST
Room 203

Workshop

10:30 CEST

Coffee break

Tuesday July 8, 2025 10:30 - 11:00 CEST

Break

10:40 CEST

Workshops

Tuesday July 8, 2025 10:40 - 12:00 CEST

Rooms 4, 5, 6, 9, 101, 103, 202, 203, Hall 1A, Hall 2B, Hall 3B, Belvedere, Onice, Auditorium, or TBA by workshop organizers

Tuesday July 8, 2025 10:40 - 12:00 CEST
Rooms 4, 5, 6, 9, 101, 103, 202, 203, Hall 1A, Hall 2B, Hall 3B, Belvedere, Onice, Auditorium, or TBA by workshop organizers

Workshops

12:30 CEST

Lunch break

Tuesday July 8, 2025 12:30 - 14:00 CEST

Break

14:00 CEST

Keynote #4: Maurizio Mattia

Tuesday July 8, 2025 14:00 - 15:20 CEST

Auditorium - Plenary Room

Speakers

Maurizio Mattia

Tuesday July 8, 2025 14:00 - 15:20 CEST
Auditorium - Plenary Room

Keynote

15:20 CEST

Conference photo

Tuesday July 8, 2025 15:20 - 15:30 CEST

TBA

Tuesday July 8, 2025 15:20 - 15:30 CEST
TBA

Registration/organization

15:30 CEST

Coffee break

Tuesday July 8, 2025 15:30 - 16:00 CEST

Break

16:00 CEST

Member's meeting

Tuesday July 8, 2025 16:00 - 17:00 CEST

Auditorium - Plenary Room

Tuesday July 8, 2025 16:00 - 17:00 CEST
Auditorium - Plenary Room

Registration/organization

17:00 CEST

Poster session 3

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Poster session

17:00 CEST

P224: Four-compartment model of dopamine dynamics at the nigrostriatal synaptic site

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P224 Four-compartment model of dopamine dynamics at the nigrostriatal synaptic site

Alex G. O'Hare*1, 2, Catalina Vich1, 2, Jonathan E. Rubin3, 4, Timothy Verstynen3, 5

1Dept. de Matemátiques i Informática, Universitat de les Illes Balears, Palma, Illes Balears, Spain
2Institute of Applied Computing and Community Code, Palma, Illes Balears, Spain
3Center for the Neural Basis of Cognition, Pittsburgh, Pennsylvania, United States of America
4Department of Mathematics, University of Pittsburgh, Pittsburgh, Pennsylvania, United States of America
5Department of Psychology & Neuroscience Institute, Carnegie Mellon University, Pittsburgh, Pennsylvania, United States of America

*Email: alex-gwyn.o-hare@uib.cat
Introduction

The traditional model of dopamine (DA) dynamics [1] posits that the level of extrasynaptic (tonic) DA modulates the effect of phasic burst firing which occurs in the event of a reward [2]. It is supposed that tonic DA, although present in low concentrations, has the capacity to activate DA synthesis and release modulating autoreceptors of the pre-synaptic DA neuron. Taking into account this traditional model as well as recent findings which demonstrate the capacity of tonic DA to also affect both D1 and D2 postsynaptic receptors [3], we develop a biologically realistic, yet computationally efficient 4-compartment model (see Fig.1) of DA action at the synaptic site, to elucidate the impact of DA dynamics on receptor occupancy and tonic DA levels.
Methods
DA is synthesised in the terminal of the presynaptic substantia nigra pars compacta (SNc) neuron, DAter, and released into the synaptic cleft, DAsyn at a rate dependent on DAter and the membrane voltage of the SNc neuron. From the synaptic cleft, DA is occupied by D1 or D2 receptors, the quantity of which is occupied we consider as the third compartment, DAocc. DAocc impacts the excitability and plasticity of postsynaptic spiny projection neuron (SPN) which receives inputs from a cortical neuron. DA is removed from DAsyn by reuptake into DAter and via diffusion to the extrasynaptic space, DAext. DA in DAsyn affects release via autoreceptors. DA in DAext affects synthesis autoreceptors and is removed from the system via diffusion.
Results
Preliminary symbolic analysis of the system in a supposed quasi-steady-state determined by setting the rate of synthesis equal to the rate of removal from DAext and letting the firing rate of the presynaptic neuron be constant, reveals a stable system according to the Routh-Hurwitz criterion, with either damped or no oscillations. Building on prior work [4] in which we developed an STDP model for cortico-striatal plasticity, we incorporate a presynaptic SNc neuron to analyse the effect of variations in parameters (limited within ranges of empirical data and using latin hypercube sampling) of the DA model on plasticity.
Discussion
Our four-compartment model of nigrostriatal dopamine dynamics bridges the gap between purely phenomenoligcal models which lack biological realism and more complex models which take into account a high degree of biological detail and are computationally expensive, thereby providing a solution for incorporating the effect of DA on corticostriatal plasticity in large scale spiking neural networks. In particular, our model may be of utility for simulations of dopaminergic reinforcement learning, such as in n-choice tasks, and simulations of DA-related pathologies which require explicit consideration of postsynaptic receptor occupation and extrasynaptic DA levels.

Figure 1. 4-compartment model of dopamine (DA) at the synaptic site. DAter: presynaptic terminal, DAsyn: synaptic cleft, DAocc: occupied postsynaptic receptors, DAext: extrasynaptic space. Pointed arrows indicate the transfer of DA from one compartment to another, with constants ki indicating the rate of transfer. Dotted arrows denote the modulatory effect of synthesis and release modulating autoreceptors.
Acknowledgements.
References
1. https://doi.org/10.1016/0376-8716(94)01066-t
2. https://doi.org/10.1126/science.275.5306.1593
3. https://doi.org/10.1523/jneurosci.1951-19.2019
4. https://doi.org/10.1016/j.cnsns.2019.105048

Speakers

Alex O'Hare

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P225: Time-to-first-spike encoding in layered networks evokes label-specific synfire chain activity

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P225 Time-to-first-spike encoding in layered networks evokes label-specific synfire chain activity

Jonas Oberste-Frielinghaus1,2, Anno C. Kurth1, Julian Göltz3,4, Laura Kriener5,4,Junji Ito*1, Mihai A. Petrovici4, Sonja Grün1,6,7

1Institute for Advanced Simulation (IAS-6), Jülich Research Centre, Jülich, Germany
2RWTH Aachen University, Aachen, Germany
3Kirchhoff-Institute for Physics, Heidelberg University, Heidelberg, Germany
4Department of Physiology, University of Bern, Bern, Switzerland
5Institute of Neuroinformatics, University of Zurich and ETH Zurich, Zurich, Switzerland
6JARA Brain Institute I (INM-10), Jülich Research Centre, Jülich, Germany
6Theoretical Systems Neurobiology, RWTH Aachen University, Aachen, Germany

*Email: j.ito@fz-juelich.de
Introduction

While artificial neural networks (ANNs) have achieved remarkable success in various tasks, they lack two major characteristic features of biological neural networks: spiking activity and operation in continuous time. This makes it difficult to leverage knowledge about ANNs to gain insights into the computational principles of the real brains. However, training methods for spiking neural networks (SNNs) have recently been developed to create functional SNN models [1]. In this study we analyze the activity of a multilayer feedforward SNN trained for image classification and uncover the structures in both connectivity and dynamics that underlie its functional performance.

Methods
Our network is composed of an input layer (784 neurons), 4 hidden layers (300 excitatory and 100 inhibitory neurons in each layer), and an output layer (10 neurons). We trained it with backpropagation to classify the MNIST dataset, based on time-to-first-spike coding: each neuron encodes information in the timing of its first spike; the first neuron to spike in the output layer defines the inferred input image class [1]. The MNIST input is also provided as spike timing: dark pixels spike early, lighter pixels later. Based on the connection weights after training, neurons that have strong excitatory effects on each of the output neurons are identified in each layer. Note that one neuron can have strong effects on multiple output neurons.
Results
In response to a sample, the input layer generates a volley of spikes, identified as a pulse packet (PP) [2], which propagates through the hidden layers (Fig. 1). In deeper layers, spikes in a PP get more synchronized and the neurons providing spikes to the PP become more specific to the sample label. This leads to a characteristic sparse representation of the sample label in deep layers. The analysis of connection weights reveals that a correct classification is achieved by propagating spikes through a specific pathway across layers, composed of neurons with strong excitatory effects on the correct output neuron. Pathways for different output neurons become more separate in deeper layers, with less overlap of neurons between pathways.
Discussion
The revealed connectivity structure and the propagation of spikes as a PP agree with the notion of the synfire chain (SFC) [3,4]. To our knowledge, this is the first example of SFC formation by training of a functional network. In our network, multiple parallel SFCs emerge through the training for MNIST classification, representing each input label by activation of one particular SFC. Such a representation naturally leads to sparser encoding of the input label in deeper layers, and also increases the linear separability of layer-wise activity. Thus, the use of SFCs for information representation can have multiple advantages for achieving efficient computation, besides the stable transmission of information through the network.

Figure 1. Network activity in response to six different samples. Dots represent spike times of individual neurons, with colors indicating the luminance of the corresponding pixels in the sample (“input” layer), or spikes of excitatory (red) and inhibitory (blue) neurons (layers 1-4). The first neurons to spike in the “output” layer are indicated by numbers next to the spikes.
Acknowledgements
This research was funded by the European Union’s Horizon 2020 Framework programme for Research and Innovation under Speciﬁc Grant Agreements No. 785907 (HBP SGA2), No. 945539 (HBP SGA3) and No. 101147319 (EBRAINS 2.0), the NRW-network 'iBehave' (NW21-049), the Helmholtz Joint Lab SMHB, and the Manfred Stärk Foundation.

References
● Göltz et al. (2021). Fast and energy-efficient neuromorphic deep learning with first-spike times. Nature Machine Intelligence, 3(9), 823–835. https://doi.org/10.1038/s42256-021-00388-x
● Diesmann, Gewaltig, & Aertsen (1999). Stable propagation of synchronous spiking in cortical neural networks. Nature, 402(6761), 529–533. https://doi.org/10.1038/990101
● Abeles (1982). Local Cortical Circuits: An Electrophysiological Study. Springer-Verlag.
● Abeles (1991). Corticonics: Neural Circuits of the Cerebral Cortex. Cambridge University Press.

Speakers

Mihai A. Petrovici

Junji Ito

Sonja Gruen

Julian Goeltz

Laura Kriener

Jonas Oberste-Frielinghaus

Anno C. Kurth

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P226: Astrocyte modulation of neural oscillations: mechanisms underlying slow wave activity in cortical networks

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P226 Astrocyte modulation of neural oscillations: mechanisms underlying slow wave activity in cortical networks

Thiago Ohno Bezerra*1, Antonio C. Roque1

1Department of Physics, School of Philosophy, Sciences and Letters of Ribeirão Preto, University of São Paulo, Ribeirão Preto, São Paulo, Brazil.

*Email: thiagotakechi@usp.br

Introduction
Oscillatory activity plays a pivotal role in neural networks. Astrocytes have recently been shown to modulate neural activity through the release of glutamate and ATP, the latter acting as an inhibitory neuromodulator, and have been implicated in the regulation of slow cortical oscillations, specifically the up and down states observed in these networks. However, the mechanisms by which astrocytes influence neural oscillations and shape network activity remain poorly understood.

Methods
We extended the INEXA model [1] to incorporate the adaptation of neural activity. Neurons (N = 250) and astrocytes (N = 75, 30% of neurons) are randomly distributed in a 3D volume (750 × 750 × 10 µm3). Each neuron is modeled as a stochastic unit, where its spiking probability depends on neural excitatory and inhibitory inputs, ATP-mediated inhibition from astrocytes, an adaptive variable (u), and background noise (c = 0.03). The variable u increases after each neuronal spike and decays over time. Presynaptic neuron activity enhances IP3concentration in astrocytes, which elevates local Ca2+levels. Astrocyte activity is modeled as a stochastic process, driven by local Ca2+responses and the activation of neighboring astrocytes. Glutamate release from astrocytes promotes synaptic facilitation, influencing neuron-to-neuron communication. Connectivity between neurons and astrocytes is governed by a probabilistic rule based on spatial proximity.

Results
The model predicts that without astrocytes, neural networks oscillate at frequencies that vary according to the increment and decay rates of the variable u. These oscillations show no slow-wave patterns. In contrast, when astrocytes are included, the network exhibits three distinct activity modes: (1) high-frequency asynchronous spiking, (2) alternating between high-frequency spiking and silent states, and (3) regular synchronous spiking. The second mode, characterized by alternating states, is particularly reminiscent of cortical up and down states associated with slow oscillations. The specific mode of activity is influenced by the dynamics of the adaptive variable u, which modulates the frequency and pattern of oscillations. Astrocytic synaptic potentiation, ATP-mediated inhibition, and astrocyte activation duration also regulate the slow oscillation frequency.

Discussion
Our results suggest that astrocytes play an integral role in modulating the activity patterns of neural networks. Through the release of glutamate and ATP, astrocytes influence both excitatory and inhibitory processes, thereby altering network dynamics. These findings support the hypothesis that astrocytes are essential for the generation and regulation of slow oscillations in cortical networks, specifically in the context of up and down states. The modulation of these oscillations by astrocytic activity may provide a mechanism through which astrocytes influence cognitive processes associated with neural synchrony.

Acknowledgements
This work was produced as part of the activities of FAPESP Research, Innovation and Dissemination Center for Neuromathematics (grant 2013/07699-0). TOB is supported by a FAPESP PhD scholarship (grant 2021/12832-7, BEPE: 2024/14422-9). ACR is partially supported by a CNPq fellowship (grant 303359/2022-6).
References
[1] Lenk, K., Satuvuori, E., Lallouette, J., Ladrón-de-Guevara, A., Berry, H., & Hyttinen, J. A. (2020). A computational model of interactions between neuronal and astrocytic networks: The role of astrocytes in the stability of the neuronal firing rate. Frontiers in Computational Neuroscience, 13, 92. https://doi.org/10.3389/fncom.2019.00092

Speakers

Thiago Ohno Bezerra

Antonio Carlos Roque

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P227: Only a matter of time: developmental heterochronicity captures network properties of the human connectome

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P227 Only a matter of time: developmental heterochronicity captures network properties of the human connectome

Stuart Oldham*1,Francesco Poli2, Duncan Astle2, Gareth Ball1

1 Murdoch Children’s Research Institute, Melbourne, Australia
2 Cambridge University, Cambridge, UK

*Email: stuart.oldham@mcri.edu.au

Brain network organization is shaped by a trade-off between connection costs and functional benefits [1]. Computational generative network models have found this trade-off explains many, but not all, network properties [2,3]. During gestation, brain development proceeds according to spatiotemporal patterns defined by morphogen gradients [4]. Cortical areas display heterochronicity, differential timing of key developmental events, which induces spatial patterns that persist in later life as smoothly varying gradients in cytoarchitecture, neuronal connectivity, and functional activation [4,5]. Therefore, we developed a new computational model to assess how heterochronicity may constrain the formation of cortical connectivity.
Developmental timing was modeled along a unimodal gradient, originating from one node per hemisphere (Fig. 1A). Nodes were sequentially 'activated' over model timesteps based on their geodesic distance from the origin, with the timing/heterochronicity of activation controlled by the parameter τ, and the connection probability between active nodes governed by their wiring-cost η (Fig. 1B-C). The summed probabilities across timesteps used to generate a density matched network (Fig. 1C). The model was run for each origin, optimizing parameters to maximize model fit, which was the degree correlation ρ with empirical network (a group consensus structural connectivity brain network [2]), a feature generative models struggle to capture [2,3].
Spatial gradients modeling heterochronicity from the frontal cortex yielded the highest degree correlations (max ρ=0.39). These same networks with the best degree correlation also captured key empirical topological features, including clustering, connection length, and primary connectivity gradient. However, they did not fully replicate modularity or connection overlap (Fig. 1D). Models from the best origins (ρ>0.25) outperformed previous leading approaches[2,3] (Fig. 1E) and achieved the best fits with strong heterochronicity (τ > 0.5) and minimal distance penalties (η ≈ 0; Fig. 1F). Models using only the heterochronicity term produced similar degree correlations (Fig. 1G), suggesting it alone can drive brain-like connectivity patterns.
Here we demonstrate that constraining network connections to form along an anterior-posterior gradient is sufficient to capture topographical and topological connectomic features of empirical brain networks. These models also outperform past approaches [2,3]. The best-performing models imposed a heterochronous gradient that aligned with the rostral-caudal axis, a known major neurodevelopmental gradient[4,5], suggesting that early spatiotemporal patterning along this axis is key to shaping cortical connectivity. While our study examined single unimodal gradients, future studies could integrate multiple biologically informed gradients to better model network complexity. Our framework offers a flexible foundation for such extended work.

Figure 1. Fig. 1 (A) Geodesic distances from example origin (B) Heterochronicity/wiring-cost calculation (C) Model connection probability and network generation (D) Similarity to the empirical data on network features for each origin’s best fitting model (E) Comparison to previous models[2] (F) τ and η for each origin’s best fitting model (G) Best degree correlations for heterochronous-only models
Acknowledgements
S.O is supported by the Brain and Behavior Research Foundation (ID: 31471). G.B. was supported by the National Health and Medical Research Council (ID: 1194497). Research was supported by the Murdoch Children’s Research Institute, the Royal Children’s Hospital, Department of Paediatrics, The University of Melbourne and the Victorian Government’s Operational Infrastructure Support Program.
References
1.https://doi.org/10.1038/nrn3214
2.https://doi.org/10.1101/2024.11.18.624192
3.https://doi.org/10.1126/sciadv.abm6127
4.https://doi.org/10.1016/j.neuron.2007.10.010

5.https://doi.org/10.1016/j.tics.2017.11.002

Speakers

Stuart Oldham

Francesco Poli

Duncan Astle

Gareth Ball

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P228: Multi-scale Spiking Network Model of Human Cerebral Cortex

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P228 Multi-scale Spiking Network Model of Human Cerebral Cortex

Renan O. Shimoura*1, Jari Pronold1,2, Alexander van Meegen1,3, Mario Senden4,5, Claus C. Hilgetag6, Rembrandt Bakker1,7, Sacha J. van Albada1,3

1Institute for Advanced Simulation (IAS-6), Jülich Research Centre, Jülich, Germany
2RWTH Aachen University, Aachen, Germany
3Institute of Zoology, University of Cologne, Cologne, Germany
4Faculty of Psychology and Neuroscience, Department of Cognitive Neuroscience, Maastricht University, Maastricht, The Netherlands
5Faculty of Psychology and Neuroscience, Maastricht Brain Imaging Centre, Maastricht University, Maastricht, The Netherlands
6Institute of Computational Neuroscience, University Medical Center Eppendorf, Hamburg University, Hamburg, Germany
7Department of Mathematics and Computer Science, Eindhoven University of Technology, Eindhoven, The Netherlands

*Email: r.shimoura@fz-juelich.de
Introduction

Data-driven models at cellular resolution have been built for various brain regions, yet few exist for the human cortex. We present a comprehensive point-neuron network model of a human cortical hemisphere integrating diverse experimental data into a unified framework bridging cellular and network scales [1]. Our approach builds on a large-scale spiking network model of macaque cortex [2,3] and investigates how resting-state activity emerges in cortical networks.

Methods
We constructed a spiking network model representing one hemisphere using the Desikan-Killiany parcellation (34 areas), with each area implemented as a 1 mm² microcircuit distinguishing the cortical layers. The model aggregates data across multiple modalities, including electron microscopy for synapse density, cytoarchitecture from the von Economo atlas [4], DTI-based connectivity [5], and local connection probabilities from the Potjans-Diesmann microcircuit [6]. Human neuron morphologies [7] inform the layer-specific inter-area connectivity. The full-density model, based on leaky integrate-and-fire neurons, comprises 3.47 million neurons with 42.8 billion synapses and was simulated using the NEST simulator on the JURECA-DC supercomputer.

Results
When local and inter-area synapses have the same strength, model simulations show asynchronous irregular activity deviating from experiments in terms of spiking activity and inter-area functional connectivity. When inter-areal connections are strengthened relative to local synapses, the model reproduces both microscopic spiking statistics from human medial frontal cortex and macroscopic resting-state fMRI correlations [8]. Analysis reveals that single-spike perturbations influence network-wide activity within 50-75 ms. The ongoing activity flows primarily from parietal through occipital and temporal to frontal areas, consistent with empirical findings during visual imagery [9].

Discussion
This open-source model integrates human data across scales to investigate cortical organization and dynamics. By preserving neuron and synapse densities, it accounts for the majority of the inputs to the modeled neurons, enhancing the self-consistency compared to downscaled models. The model allows systematic study of structure-dynamics relationships and forms a platform for investigating theories of cortical function. Future work may leverage the Julich-Brain Atlas to refine the parcellation and incorporate detailed cytoarchitectural and receptor distribution data [10]. The model code is publicly available athttps://github.com/INM-6/human-multi-area-model.

Acknowledgements
This work was supported by the German Research Foundation (DFG) Priority Program "Computational Connectomics" (SPP 2041; Project 347572269), the EU Grant 945539 (HBP), EBRAINS 2.0 Project (101147319), the Joint lab SMHB, and HiRSE_PS. The use of the JURECA-DC supercomputer in Jülich was made possible through VSR computation grant JINB33. Open access publication funded by DGF Grant 491111487.
References
[1] https://doi.org/10.1093/CERCOR/BHAE409.
[2] https://doi.org/10.1007/s00429-017-1554-4.
[3] https://doi.org/10.1371/journal.pcbi.1006359.
[4] https://doi.org/10.1159/isbn.978-3-8055-9062-4.
[5]https://doi.org/10.1016/J.NEUROIMAGE.2013.05.041.
[6] https://doi.org/10.1093/cercor/bhs358.
[7] https://doi.org/10.1093/CERCOR/BHV188.
[8] https://doi.org/10.1126/science.aba3313.
[9] https://doi.org/10.1016/J.NEUROIMAGE.2014.05.081.
[10] https://doi.org/10.3389/fnana.2017.00078.

Speakers

Claus Hilgetag

Sacha van Albada

Renan Oliveira Shimoura

Jari Pronold

Alexander van Meegen

Mario Senden

Rembrandt Bakker

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P229: Exploring Electroencephalographic (EEG) Models of Brain Activity using Automated Modelling Techniques

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P229 Exploring Electroencephalographic (EEG) Models of Brain Activity using Automated Modelling Techniques

Nina Omejc*1, 2, Sabin Roman1, Ljupčo Todorovski1,3, Sašo Džeroski1

1Department of Knowledge Technologies, Jozef Stefan Institute, Ljubljana, Slovenia
2Jozef Stefan International Postgraduate School, Ljubljana, Slovenia
3Department of Mathematics, Faculty of Mathematics and Physics, Ljubljana, Slovenia

*Email: nina.omejc@ijs.si
Introduction

Electroencephalography (EEG) is a clinical, non-invasive, high-temporal resolution technique for measuring whole-brain activity. However, the underlying mechanisms that give rise to the observed high-level rhythmic activity remain incompletely understood. Various neural population and network models attempt to explain these dynamics [1], but, to our knowledge, they have not been systematically explored or evaluated.
Methods
To explore the space of proposed and potential models, we represent brain networks as graphs, where nodes correspond to brain sources obtained via EEG source analysis, in our case the dipole fitted independent components (Figure 1). Each node’s dynamics are further categorized into three subdynamics: synapto-dendritic dynamics (input transformation), intrinsic dynamic, and firing response (output transformation). These subdynamics are defined by a bounded set of functions derived from the literature [1], or generated by an unbounded probabilistic context-free grammar [2]. Such a modular and unbounded specification allows for flexible and physiologically valid construction of the network.
Results
We are currently utilizing our Julia-based framework and are in the model evaluation phase. The dataset we use consists of 64-channel EEG recordings from 50 participants performing a visual flickering task, designed to induce steady-state visual evoked potentials [3]. We repeatedly sample potential EEG models using Markov Chain Monte Carlo and optimize the model parameters using CMA-ES algorithm. By the time of the conference, we aim to determine which established and previously unexamined whole-brain activity models can reproduce the observed oscillations, and, more importantly, which can also accurately capture the harmonics of the flickering stimulation frequency, a robust and interesting feature observed in this dataset.
Discussion
The presence of these harmonic components is a well-documented but not yet fully understood phenomenon in EEG research [4]. By systematically exploring different model configurations, we aim to assess which types of nonlinear models and which features (for example, recurrent connectivity, nonlinear synaptic integration, parallel computations, delays) play a crucial role in shaping these spectral patterns. Exploring the set of valid models to understand these mechanisms could have broader implications for theories of whole-brain neural activity and improve our understanding of EEG measurements.

Figure 1. Figure 1: A data-driven framework for exploring whole-brain network EEG models.
Acknowledgements
We would like to thank our department's SHED group for equation discovery for the fruitful discussions regarding our work.
References

[1]https://doi.org/10.1007/978-3-030-89439-9_13
[2]https://doi.org/10.1007/s10994-024-06522-1
[3]https://doi.org/10.1093/gigascience/giz002
[4]https://doi.org/10.1016/j.neuroimage.2012.05.054

Speakers

Nina Omejc

Sabin Roman

Ljupčo Todorovski

Sašo Džeroski

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P230: Driver Nodes for Efficient Activity Propagation Between Clusters in Spiking Neural Networks

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P230 Driver Nodes for Efficient Activity Propagation Between Clusters in Spiking Neural Networks

Bulat Batuev1+,Arsenii Onuchin2,3+, Sergey Sukhov1

1Kotelnikov Institute of Radioengineering and Electronics of Russian Academy of Sciences, Moscow, Russia
2Skolkovo Institute of Science and Technology, Moscow, Russia
3Laboratory of Complex Networks, Center for Neurophysics and Neuromorphic Technologies, Moscow, Russia

+ These authors contributed equally

Email: arseniyonuchin04.09.97@gmail.com

Introduction

Synchronous neural activity is critical for brain function, yet the connectome's role in enabling synchronization remains unclear. We explore strategies to achieve widespread synchronization in spiking stochastic block model (SBM) networks with minimal control inputs. This work builds on research into neural network control [1], focusing on identifying driver nodes that influence dynamics. By evaluating centrality measures (betweenness, degree, eigenvector, closeness, harmonic, percolation), we pinpoint topological features predicting effective driver nodes. Furthermore, we analyze connectivity patterns to understand pairwise activity relationships and uncover mechanisms of network-wide coordination.

Methods
The spiking neural network consisted of 500 leaky integrate-and-fire neurons (80% excitatory, 20% inhibitory) divided into two clusters of equal size, with intra-cluster edge probability 0.15 and inter-cluster varying from 0.06 to 0.13. To simulate background activity, all neurons received independent Poisson-distributed inputs. Within the first cluster, a subpopulation of neurons (10–20%) was designated as driver neurons and subjected to an additional external current stimulus (10 Hz, 1000 pA). Driver neurons were selected either randomly or by centrality metrics (betweenness, degree, eigenvector, closeness, harmonic, percolation) [2]. Neural dynamics were simulated for 5 seconds to achieve steady-state activity using the Brian 2 [3].
Results
The population activity in the non-stimulated cluster was analyzed as a function of the number of driver neurons and the inter-cluster connectivity. When driver neurons were selected using closeness and betweenness centrality metrics, spike rates in the second cluster increased approximately 10-fold compared to random selection, accompanied by synchronization with the first cluster at nearly 10 Hz. In contrast, selecting driver neurons based on degree and percolation centrality metrics resulted only in 5-fold increase compared to random selection (Fig. 1).

Discussion
Synchronization between two weakly coupled clusters can be achieved by selectively stimulating specific neurons within the first cluster. However, it remains unclear why closeness and betweenness centrality outperform other centralities in promoting synchronization. Future research could focus on extending our method to multicluster heterogeneous systems. While the two-cluster model offers a controlled setting, expanding it could provide deeper insights into real brain connectomes. In conclusion, this study elucidates how topology and driver node selection shape neural synchronization, with potential applications in neuromodulation and brain-inspired systems.

Figure 1. The average population activity within the second cluster, calculated over a 1-second time window, is depicted for driver nodes selected based on various centrality measures (degree, betweenness, eigenvector centrality, PageRank, and percolation centrality) for the upper surface, and for nodes chosen at random for the lower surface.
Acknowledgements

This work was funded by the Russian Science Foundation (project number 24-21-00470).
References
1.Bayati, M., Valizadeh, A., Abbassian, A., & Cheng, S. (2015). Self-organization of synchronous activity propagation in neuronal networks driven by local excitation.Frontiers in Computational Neuroscience, 9, 69. https://doi.org/10.3389/fncom.2015.00069
2.Saxena, A., & Iyengar, S. (2020). Centrality measures in complex networks: A survey. arXiv preprint arXiv:2011.07190.
3.Stimberg, M., Brette, R., & Goodman, D. F. (2019). Brian 2, an intuitive and efficient neural simulator.eLife, 8, e47314. https://doi.org/10.7554/eLife.47314

Speakers

Arsenii Onuchin

Sergey Sukhov

Bulat Batuev

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P231: Emotional network modelling: whole brain simulations of fear conditioning in humans

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P231 Emotional network modelling: whole brain simulations of fear conditioning in humans

Dianela A Osorio-Becerra1, Andrea Fusari1, Ashika Roy2, Danilo Benozzo1, Andreas Frick2, Egidio D’Angelo1, Fulvia Palesi1,Claudia Casellato1
1Department of Brain and Behavioral Sciences, University of Pavia, Pavia, Italy
2Department of Medical Sciences, Uppsala University, Uppsala, Sweden
p { line-height: 115%; text-align: left; orphans: 2; widows: 2; margin-bottom: 0.1in; direction: ltr; background: transparent }a:link { color: #467886; text-decoration: underline }

*Email: claudia.casellato@unipv.it
Introduction

Emotion in mammals involves complex brain networks [1,2], for which it is critical to identify the specific regional connectivity and microcircuit properties. We couple in-silico brain dynamics in whole-brain simulations by The Virtual Brain – TVB [3] with experimental fear conditioning data in humans. These include MRI (DWI, resting-state and task-dependent fMRI) and fear-related behavioural measurements (skin conductance responses (SCR), a biomarker of emotional arousal [4]). This work represents a preliminary exploration on how data-driven subject-specific models of brain dynamics could predict emotional behaviour.
Methods
Data come from 17 healthy subjects. The fMRI is acquired at TR 3 s. The mean SCR is extracted for CS+ (conditioned stimulus paired with the unconditioned stimulus, US) and CS− (neutral stimulus), both during acquisition and extinction (acq_csp, acq_csm, ext_csp, ext_csm), and for US (electric shock) during acquisition (acq_us). The SCR usually increases with the paired CS/US pattern presentation, and it decreases when CS is no longer followed by US (extinction).
In the model, a fear-TVB network was defined selecting the regions involved, with each node represented by a reduced Wong-Wang model [5] and using the subject-specific structural connectivity from DWI. Then, the fear-TVB network was optimized in terms of global and local connection parameters, by maximizing the match between the subject-specific experimental functional connectivity matrices (static and dynamic – expFC and expFCD), obtained from resting-state fMRI, and the simulated ones (simFC and simFCD). Finally, these parameters were correlated with the subject-specific SCRs.
Results
The fear-TVB network was reconstructed using 88 nodes, including the amygdala, cerebellum, periaqueductal gray, and parts of the limbic system. The TVB parameters - i.e. global couplingGand three synaptic parameters (excitatory NMDA strengthJNMDA, inhibitory GABA strengthJi, recurrent excitationw+) - were extracted during the optimization process, see Fig.1. By correlating TVB parameters with SCR measures, a positive correlation betweenGandacq_us(ρ=0.37)emerged. However, higher correlation was found when considering sex separately, reinforcing the existing literature on this field.

Discussion
These findings suggest that individual differences in resting-state neural dynamics influence fear acquisition, with distinct mechanisms supporting US processing and conditioned fear discrimination. Although it is the first time that a correlation between network dynamics and fear responses is revealed, the relationship between global connectivity strength and fear responses is still weak, more data and closer understanding of the underlying network is needed. The next step is to use the fMRI data along fear conditioning trials by defining a time-dependent subject-specific TVB parameter space, which may correlate with the corresponding time-dependent fear responses.

Figure 1. a)Fear network b)Anterior and posterior views, 88 nodes (frontal, prefrontal, limbic, parietal, temporal, occipital, deep ganglia, brainstem and cerebellum) c)Violin plots of fear-measures and TVB parameters d)Exp and sim FC matrices, mean across subjects, each element is the Pearson Correlation Coefficient e)Exp vs sim: PCC for FC and the Kolmogorov–Smirnov distance for FCD, one point for subject
Acknowledgements
European Union's Horizon 2020 research under the Marie Sklodowska-Curie grant agreement No. 956414 for "Cerebellum and Emotional Networks", and #NEXTGENERATIONEU, by the Ministry of University and Research, National Recovery and Resilience Plan, project MNESYS (PE0000006)-A Multiscale integrated approach to the study of the nervous system in health and disease (DN. 1553 11.10.2022).
References
[1]https://doi.org/10.3389/fnsys.2023.1185752
[2]https://doi.org/10.1146/annurev.neuro.23.1.155
[3]https://doi.org/10.3389/fninf.2013.00010
[4]https://doi.org/10.1177/1094428116681073[5]https://doi.org/10.1523/JNEUROSCI.5068-13.201

Speakers

Claudia Casellato

Danilo Benozzo

Egidio D'Angelo

Dianela Andreina Osorio-Becerra

Andrea Fusari

Ashika Roy

Andreas Frick

Fulvia Palesi

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P232: Centralized brain networks underlie grooming body part coordination

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P232 Centralized brain networks underlie grooming body part coordination

Pembe Gizem Ozdil*1,2,Clara Scherrer1, Jonathan Arreguit2, Auke Ijspeert2, Pavan Ramdya1

1Neuroengineering Laboratory, Brain Mind Institute & Interfaculty Institute of Bioengineering, EPFL, Lausanne, Switzerland
2Biorobotics Laboratory, Institute of Bioengineering, EPFL, Lausanne, Switzerland

*Email: pembe.ozdil@epfl.ch
Introduction

Animals must coordinate multiple body parts to perform essential tasks such as locomotion and grooming. While locomotor coordination has been extensively studied [1,2], less is known about how the nervous system synchronizes movements across distant body parts, such as the head and legs, to execute complex behaviors. Antennal grooming inDrosophila melanogasterprovides a powerful model to study such coordination, as flies exhibit a rich repertoire of precisely controlled limb movements. With a compact yet fully mapped nervous system,Drosophilaenables circuit-level insights into how neural networks integrate motor commands for efficient multi-limb control.

Methods

Here, we combined behavioral analyses, biomechanical modeling, and connectome-based neural circuit simulations to investigate how flies coordinate head, antennae, and forelegs during grooming. We tracked detailed movement kinematics in freely behaving flies using 3D pose estimation[3]. To understand the functional role of coordination, recorded movements were replayed in a biomechanical simulation (NeuroMechFly [4]) to measure contact forces. To test proprioceptive contributions, we performed limb amputations and head immobilizations. Lastly, we analyzed the antennal grooming network using graph-based and computational neural network simulations derived from the brain connectome [5].

Results

Flies exhibit two main grooming strategies, unilateral and bilateral antennal grooming, each requiring precise coordination of head, antennae, and forelegs. Biomechanical simulations revealed that this coordination enhances grooming efficiency by avoiding obstructions and enabling forceful limb-antennal interactions. Manipulations showed proprioceptive feedback is not necessary for body-part synchronization, implying feedforward neural control. Connectome network analyses and simulations identified centralized interneurons forming recurrent excitatory and broad inhibitory circuit motifs that robustly synchronize motor modules. We further validated some model predictions through optogenetic experiments.

Discussion

We identified centralized neural circuits underlying multi-body-part coordination during antennal grooming in flies. Unlike locomotion, where coordination often depends on sensory feedback, grooming synchronization is centrally driven, likely reducing sensory processing demands. We uncovered two neural circuit motifs—recurrent excitation promoting targeted movements, and broadcast inhibition suppressing competing actions—that enable precise yet flexible coordination. This centralized circuit architecture may represent a general neural strategy conserved across behaviors and species, simplifying motor control and facilitating the evolution of complex behaviors through modular coordination.

Acknowledgements
PR acknowledges support from an SNSF Project Grant (175667) and an SNSF Eccellenza Grant (181239). JA acknowledges support from a European Research Council Synergy grant (951477). PGO acknowledges support from a Swiss Government Excellence Scholarship for Doctoral Studies and a Google PhD Fellowship.

References
● https://doi.org/10.1016/S0959-4388(98)80114-1
● https://doi.org/10.1152/jn.00658.2017
● https://doi.org/10.1016/j.celrep.2021.109730

● https://doi.org/10.1038/s41592-022-01466-7
● https://doi.org/10.1038/s41586-024-07558-y

Speakers

Pembe Gizem Ozdil

Clara Scherrer

Jonathan Arreguit

Auke Ijspeert

Pavan Ramdya

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P233: Extracellular K+ hotspots regulate synaptic integration in the dendrites of pyramidal neurons

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P233 Extracellular K+ hotspots regulate synaptic integration in the dendrites of pyramidal neurons

Malthe S. Nordentoft1, Naoya Takahashi2, Mathias S. Heltberg1, Mogens H. Jensen1, Rune N. Rasmussen3,Athanasia Papoutsi*4
1 Niels Bohr Institute, University of Copenhagen, Copenhagen, Denmark
2 Interdisciplinary Institute for Neuroscience, University of Bordeaux, Bordeaux, France
3 Center for Translational Neuromedicine, University of Copenhagen, Copenhagen, Denmark
4 Institute of Molecular Biology and Biotechnology, Foundation for Research and Technology—Hellas, Crete, Greece

*Email:papoutsi@imbb.forth.gr

Introduction
Throughoutthenervoussystem,neuronal activity and ionic changes in the extracellular environment are bidirectionally linked. Changesintheconcentration of extracellularK+ions ([K+]o)are particularly intriguing due to its pivotal role in shaping neuronal excitability and its activity- and state-dependent fluctuation [1]. At the synaptic level, [K+]ochanges arise mainly from the activation of NMDA receptors and are highly localized [2]. Despite this experimental evidence, local, activity-dependent [K+]ochanges have not been considered an integral part of neuronal signaling. In this work [3], we hypothesize that [K+]ochanges form “K+hotspots” that locally regulate the active dendritic properties and shape sensory processing.
Methods
We focus on the organization of orientation-tuned synapses on dendrites of visual cortex pyramidal neurons [4], as we have previously shown that visual cortex responses are dynamically regulated by [K+]oand brain states [1]. We first analytically investigate the spatial diffusion of K+ions, to evaluate the creation of“K+hotspots”. Following, by treating orientation-tuned inputs to dendritic segments as statistical ensembles, we infer the expected changes in Δ[K+]oand the correspondingEK+shifts. Finally, using biophysically realistic models of a point dendrite anda morphologically detailed neuron, weevaluate theeffect of the differentEK+shifts in thedendritic spike propertiesand the neuronal output.
Results
Our statistical approach identified the expectedEK+shifts under different extracellular space sizes, intracellularK+concentration changes, and presented stimuli. Importantly, dendritic segments receiving similarly-tuned inputs attain substantially higher [K+]oandEK+shifts, with theEK+shifts being within the 6-18mV range. In the point dendrite model, this range ofEK+shifts broadens dendritic spikes and increases dendritic spike probability. Finally, in the morphologically detailed neuron models, we show that the local activity-dependent[K+]oincrease andEK+shifts in dendrites enhance the effectiveness of distal synaptic inputs to cause feature-tuned firing of neurons, without comprising feature selectivity.
Discussion
In this work [3] we show that dendrites receiving similarly-tuned inputs support activity-dependent, local changes in[K+]o, forming “K+hotspots”. These hotspots depolarizeEK+and increase the reliability and duration of dendritic spikes. These effects act as a volume knob of dendritic input, promoting gain amplification of neuronal output without affecting the feature selectivity. Overall, compared to long-term plasticity mechanisms, “K+hotspots” are transient, closely follow the overall dendritic activity levels and selectively boost integration of synaptic inputs with minimum usage of resources. Our results therefore suggest a prominent and previously overlooked role of [K+]ochanges.

Acknowledgements
We thank Akihiro Matsumoto, Alessandra Lucchetti, Eva Maria Meier Carlsen, Ioannis Matthaiakakis, and Stamatios Aliprantis for discussions and comments on this work.
References

1.https://doi.org/10.1016/j.celrep.2019.06.082
2.https://doi.org/10.1016/j.celrep.2013.10.026
3.https://doi.org/10.1371/journal.pbio.3002935*PLoS Biology Issue Image | Vol. 22(12) January 2025
4.https://doi.org/10.1038/s41467-019-13029-0

Speakers

Athanasia Papoutsi

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P234: Complexity of an astrocyte-neuron network model in random and hub-driven connectivity

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P234 Complexity of an astrocyte-neuron network model in random and hub-driven connectivity

Paolo Paradisi*1,2,Giulia Salzano3,Marco Cafiso1,4, Enrico Cataldo4

1ISTI-CNR-Institute of Information Science and Technologies “A. Faedo”, Pisa, Italy

2BCAM-Basque Center for Applied Mathematics, Bilbao, Basque Country, Spain

3Department of Neuroscience, International School for Advanced Studies, Trieste, Italy

4Department of Physics, University of Pisa, Pisa, Italy

Introduction

The role of glial cells, particularly astrocytes, in brain neural networks has been historically

overlooked due to a neuron-centric perspective. Recent research highlights astrocytes’

involvement in synaptic modulation, memory formation, and neural synchronization, leading to

their inclusion in mathematical brain models. Concurrently, network topology plays a critical role

inneural function, with models such as random and scale-free networks offering insights into

connectivity patterns. In this work we present the investigation of a recently published astrocyte-

neuron network model [1,2], hereafter named SGBD model, consisting of excitatory and

inhibitory

leaky-integrate and fire (LIF) neural models endowed with astrocytes, activated by synaptic

transmission and modulating

Methods
Firstly, a modified version of the model is proposed in order to overcome the limitations of the SGBD model by incorporating biologically plausible features that are more compatible with the experimental results, in particular with regard to the spatial distribution of inhibitory neurons, astrocyte dynamics such as to trigger more realistic calcium oscillations, and neuron-astrocyte connections that are more intuitively linked to their spatial positioning. Then, the role of neuron-neuron connectivity is investigated by comparing randomandhub-driven connectivitiesinboth incoming and outcoming connections.Simulations are implemented using the Brian2 simulator, allowing for a comparative analysis of neural network activity with and without astrocytes.

Results
The proposed modifications lead to a more biologically realistic representation, influencing firing rates and inter-spike interval distributions. Comparisons between random and hub-driven connectivity highlight differences in network efficiency, in particular firing activity is much larger for hub-driven connectivity even if the number of links is much lower with respect to the random connectivity. Temporal complexity of avalanches is investigated through intermittency-driven complexity tools [3,4] and significant differences are found when comparing both random vs. hub-driven and astrocyte vs. no-astrocyte.

Discussion
This study reinforces the importance of astrocytes in neural network modeling and demonstrates how connectivity patterns impact temporal complexity of firing patterns. Hub-driven degree distribution is not strictly scale-free, i.e., does not display power-law decay, but, despite this, hub-driven topology triggers the emergence of power-law behavior in the inter-spike time distributions that does not emerge in the random connectivity. Similar findings are seen in the temporal complexity of neural avalanches, where different regimes of power-law scaling behavior are found.

Acknowledgements
This work was supported by the Next-Generation-EU programme under the funding schemes

PNRR-PE-AI scheme (M4C2, investment 1.3, line on AI) FAIR “Future Artificial Intelligence

Research”, grant id PE00000013, Spoke-8: Pervasive AI.

References

[1] M. Stimberg et al. (2019), Brian 2, an intuitive and efficient neural simulator, elife8, e47314.

doi:10.7554/eLife.47314

[2] M. Stimberg et al. (2019), Modeling Neuron–Glia Interactions with the Brian 2 Simulator,

Springer, Cham, 471–505. doi:10.1007/978-3-030-00817-8_18

[3] P. Paradisi, P. Allegrini, Intermittency-driven complexity in signal processing (2017), Springer,

Cham, 161–195. doi: 10.1007/978-3-319-58709-7_6

[4] P. Paradisi et al., The emergence of self-organization in complex systems-Preface (2015),

Chaos Sol. Fract.81b, 407-411. doi: 10.1016/j.chaos.2015.09.017

Speakers

Marco Cafiso

Speakers

Tanel Peets

Kert Tamm

Jüri Engelbrecht

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P238: Identifying cortical learning algorithms using brain-machine interfaces

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P238 Identifying cortical learning algorithms using brain-machine interfaces

Sofia Pereira da Silva1,2, Denis Alevi1, Friedrich Schuessler*1,3, Henning Sprekeler*1,2,3

1 Modelling of Cognitive Processes, Technische Universität Berlin, Berlin, Germany
2 Bernstein Center for Computational Neuroscience Berlin, Berlin, Germany
3 Science of Intelligence, Technische Universität Berlin, Berlin, Germany

Email: sofia@bccn-berlin.de
Introduction

By causally mapping neural activity to behavior [1], brain–machine interfaces (BMI) offer a means to study the dynamics of sensorimotor learning. Here, we investigate the neural learning algorithm monkeys use to adapt to a changed output mapping in a center-out reaching task [2]. We exploit that the mapping from neural space (ca. 100 dimensions) to the 2D cursor position is a credit assignment problem [3] that is underconstrained, because changes along a large number of output-null dimensions do not influence the behavioral output. We hypothesized that different, but equally performing learning algorithms can be distinguished by the changes they generate in output-null dimensions.

Methods
We combine computational modeling and data analysis to study the neural algorithms underlying learning in the BMI center—out task. We implement networks with three different learning rules (gradient descent, model-based feedback alignment, and reinforcement learning) and three distinct learning strategies (direct, re–aiming [4], remodeling [5]) in feedforward and recurrent architectures. The models’ initial conditions are constrained using publicly available data from BMI experiments [6, 7,8]. We train the models in cursor space and use linear regression to compare the resulting changes in neural space to the data.

Results
We first verify that all implemented algorithms can learn the task in cursor space. In terms of neural activity, we find that various combinations of rules and architectures lead to changes in different low–dimensional subspaces. For instance, re-aiming is, by definition, constrained to a lower-dimensional subspace, so the neural activity changes across algorithms within this strategy are more similar than those in other strategies. Comparing the changes in neural activity and their subspaces with available data from BMI experiments points to learning as a combination of different algorithms. However, not all variance is explained by the algorithms, indicating additional changes outside the modeled subspaces.

Discussion
Bridging BMI experiments and population dynamics analyses creates a framework to study how learning unfolds in the brain. Our results suggest that monkeys employ a combination of previously suggested strategies to learn BMI tasks, involving both model-based and model-free learning. Future work should explore models with recurrent architectures further to better capture biological dynamics. Moreover, applying methods that describe the learning manifolds and trial-to-trial variability could offer interesting insights for comparing the models and data. Finally, comparing our findings with longitudinal datasets that monitor the learning process over time would be valuable for understanding how the learning dynamics progress.

Acknowledgements

References
1.https://doi.org/10.1016/j.conb.2015.12.005
2.https://proceedings.neurips.cc/paper_files/paper/2022/hash/a6d94c38506f16fb50894a5b555f2c9a-Abstract-Conference.html
3.https://doi.org/10.1371/journal.pcbi.1008621
4.https://doi.org/10.1101/2024.04.18.589952
5.https://doi.org/10.7554/eLife.10015
6.https://doi.org/10.1038/s41593-018-0095-3
7.https://doi.org/10.1038/s41593-021-00822-8
8.https://doi.org/10.7554/eLife.36774

Speakers

Denis Alevi

Henning Sprekeler

Speakers

Dimitrios Pinotsis

Earl K. Miller

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P243: Hierarchical fluctuations scales in whole-brain resting activity

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P243 Hierarchical fluctuations scales in whole-brain resting activity

Adrián Ponce-Alvarez1,2,3*

1Departament de Matemàtiques, Universitat Politècnica de Catalunya, Barcelona, Spain.
2Institut de Matemàtiques de la UPC - Barcelona Tech (IMTech), Barcelona, Spain.
3Centre de Recerca Matemàtica, Barcelona, Spain.

*Email :adrian.ponce@upc.edu
Introduction

Brain activity fluctuates at different timescales across regions, with higher-order areas exhibiting slower dynamics than sensory regions [1]. Connectivity and local properties shape this hierarchy: spine density and synaptic gene expression gradients correlate with timescales [2–4], while strongly connected regions exhibit slower dynamics [5].
Beyond temporal features, signal variability has been linked to aging [6], brain states [7], disorders [8], and tasks [9]. However, whether spontaneous activity variance is hierarchically organized remains unknown.
This work analyses the relation between timescales, variances, and connectivity using human f/dMRI data, while exploring the mechanisms through connectome-based whole-brain models.
Methods
Publicly available data from the Human Connectome Project was used, consisting of connectome matrices and resting-state (rs) fMRI signals from 100 subjects across 3 parcellations. For each ROI, the average variance of the rs-fMRI signal, the node’s strength of the connectome, and the autocorrelation function (ACF) were calculated.

To model the variance and temporal scales of resting-state fluctuations, two commonly used whole-brain models were studied here, namely the Hopf and the Wilson-Cowan models. These models use the brain’s connectome to coupled local nodes displaying noise-driven oscillations, with intrinsic dynamics either homogeneous or constrained by the T1w/T2w macroscopic gradient.
Results
Results show that while more connected brain regions have longer timescales, their activity fluctuations exhibit lower variance. Using the Hopf and Wilson-Cowan models, we found that variance and timescales can oppositely relate to connectivity within specific model’s parameter regions, even when all nodes have the same intrinsic dynamics —but also when intrinsic dynamics are constrained by the myelinization-related macroscopic gradient. These findings suggest that connectivity and network state alone can explain regional differences in fluctuation scales. Ultimately, timescale and variance hierarchies reflect a balance between stability and responsivity, with faster, greater responsiveness at the periphery and robustness at the core.
Discussion
This study shows that the variance of fluctuations is hierarchically organized but, in contrast to timescales, it decreases with structural connectivity. Whole-brain models show that the hierarchies of timescales and variances jointly emerge within specific parameter regions, indicating a state-dependence that could serve as a biomarker for different behavioral, vigilance, or conscious states, and neuropsychiatric disorders Finally, in line with previous works on principles of core-periphery network structures [10–12], these hierarchies link to the responsivity of different network parts, with greater and faster responsiveness at the network periphery and more stable dynamics at the core, achieving a balance between stability and responsiveness.

Acknowledgements
A.P-A. is supported by the Ramón y Cajal Grant RYC2020-029117-I funded by MICIU/AEI/10.13039/501100011033 and "ESF Investing in your future". This work is supported by the Spanish State Research Agency, through the Severo Ochoa and María de Maeztu Program for Centers and Units of Excellence in R&D (CEX2020-001084-M).
References
1.https://doi.org/10.1038/nn.3862
2.https://doi.org/10.1093/cercor/bhg093
3.https://doi.org/10.1016/j.neuron.2015.09.008
4.https://doi.org/10.1038/s41593-018-0195-0
5.https://doi.org/10.1162/netn_a_00151
6.https://doi.org/10.1523/JNEUROSCI.5641-10.2011
7.https://doi.org/10.1098/rsif.2013.0048
8.https://doi.org/10.1371/journal.pcbi.1012692
9.https://doi.org/10.1523/JNEUROSCI.2922-12.2013
10.https://doi.org/10.1038/nrg1471
11.https://doi.org/10.1093/comnet/cnt016
12.https://doi.org/10.1098/rstb.2014.0165

Speakers

Adrián Ponce-Alvarez

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P244: Multi-network Modeling of Parkinson’s Disease: Bridging Dopaminergic Modulation and Vibrotactile Coordinated Reset Therapy

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P244 Multi-network Modeling of Parkinson’s Disease: Bridging Dopaminergic Modulation and Vibrotactile Coordinated Reset Therapy

Mariia Popova*1, Fatemeh Sadeghi2, Simone Zittel2, Claus C Hilgetag1,3

1Institute of Computational Neuroscience, Hamburg Center of Neuroscience, University Medical Center Hamburg-Eppendorf (UKE), Hamburg University, Hamburg, Germany
2Department of Neurology, University Medical Center Hamburg-Eppendorf, Hamburg, Germany
3Center for Biomedical AI, University Medical Center Hamburg-Eppendorf, Hamburg, Germany

*Email: m.popova@uke.de
Introduction

Computational models of Parkinson’s disease (PD) play an important role in understanding the complex neural mechanisms underlying motor symptoms, such as tremor, and in assessing novel treatment interventions. According to the “finger-dimmer-switch (FDS)” theory, tremor originates within the basal ganglia–thalamo-cortical (BTC) network, but subsequently spreads to the cerebello–thalamo-cortical (CTC) network through excessive inter-network synchronization [1]. One approach to manage severe PD tremor is to use deep brain stimulation (DBS). Recently, a new non-invasive approach of vibrotactile coordinated reset (vCR) stimulation was proposed as an alternative to DBS [2]. Here, we aimed to explore how vCR affects tremor in a computational model.
Methods
Building on the FDS, we developed a multi-network FDS model encompassing 700 neurons across 11 regions within the BTC, CTC, and thalamic networks. By modulating dopaminergic synaptic connections, we simulated the transition from a healthy state to a Parkinsonian state. Further adjustments of self-inhibition in thalamic nuclei drove tremor onset and offset.
Results
As hypothesized, dopaminergic restoration significantly reduced tremor amplitude and reinforced the thalamus as a pivotal hub for stabilizing neuronal activity. Next, we incorporated a variant of the model featuring spike-timing-dependent plasticity (STDP) to investigate vCR stimulation, a noninvasive therapy that applies patterned tactile pulses to disrupt pathological network synchronization. In line with previous theoretical findings, our simulations showed that vCR not only attenuated excessive beta-band oscillations but also unlearned maladaptive plasticity via STDP, suggesting a broader corrective effect on dysfunctional motor circuitry than dopaminergic interventions alone.
Discussion
These findings highlight the capacity of in silico models to guide therapeutic strategies, demonstrating that vCR may be of use in managing PD symptoms. Consequently, the parameter specifications of vCR should be investigated further in theoretical and clinical studies, as it may reduce patients’ reliance on pharmacological and surgical treatments.

Acknowledgements
This study was funded by the EU project euSNN (MSCAITN-ETN H2020-860563) and Deutsche Forschungsgemeinschaft (DFG, German Research Foundation)—SFB 936—Project-ID 178316478-A1/Z3.
References
[1]https://doi.org/10.1016/j.nbd.2015.10.009
[2]https://doi.org/10.4103/1673-5374.329001

Speakers

Claus Hilgetag

Mariia Popova

Fatemeh Sadeghi

Simone Zittel

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P245: Computational modeling of neural signal disruptions to predict multiple sclerosis progression

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P245 Computational modeling of neural signal disruptions to predict multiple sclerosis progression

Vishnu Prathapan*1, Peter Eipert1, Markus Kipp2, Revathi Appali3,4, Oliver Schmitt1,2
1Medical School Hamburg University of Applied Sciences and Medical University, Am Kaiserkai 1, 20457, Hamburg, Germany
2Department of Anatomy, University of Rostock Gertrudenstr 9, 18057, Rostock, Germany
3Institute of General Electrical Engineering, University of Rostock, Albert-Einstein-Straße 2, 18059, Rostock, Germany

4Department of Aging of Individuals and Society, Interdisciplinary Faculty, University of Rostock, Universitätsplatz 1, 18055, Rostock, Germany
*vishnupratapan@gmail.com

Introduction
A computational approach is proposed to overcome the limitations of existing methods in predicting Multiple Sclerosis (MS) progression. MS is marked by myelin sheath disruption, impairing neuronal signal transmission and leading to neurodegeneration and functional decline. Predicting MS progression is challenging due to disease heterogeneity, limited longitudinal data, small sample sizes, and data inconsistencies. Current models rely on static biomarkers, failing to capture dynamic interactions between immune responses, neurodegeneration, and remyelination. Furthermore, the absence of personalized models and challenges in integrating multimodal data hinder early intervention and treatment optimization [1].
Methods
This study analyzes dynamic network changes, in response to localized disturbances, offering deeper insights into MS disease progression. The Izhikevich neuron model [2] is used for its computational efficiency, scalability, and ability to simulate diverse neuronal firing patterns relevant to specific brain regions. A myelin-based delay quotient adapted based on prior research [3, 4], models demyelination and remyelination effects observed in MS. The model is validated using varied conduction values, connection weights, and nodal lengths in a three-node configuration before extending to complex networks. Finally, interconnected neuronal modules representing distinct brain regions are simulated to replicate MS conditions.
Results
Signal propagation patterns are analyzed by altering myelin-based conduction delay parameters at specific nodes, with results compared against a control model. As expected, conduction deficits significantly impact network dynamics, illustrating how neuronal signaling adapts to disease-induced disruptions.
Discussion
This model could provide insights into MS progression by capturing evolving network disruptions when applied to a connectome. This computational approach holds promise as a foundation for predictive clinical tools, supporting early diagnosis and treatment strategies. This study offers a novel perspective on MS progression and potential therapeutic interventions by integrating dynamic network modelling with biological mechanisms.

Acknowledgements
The authors thank the University of Rostock, and the Medical School Hamburg University of Applied Sciences and Medical University for institutional support.
References
1. Prathapan, V., Eipert, P., Wigger, N., Kipp, M., Appali, R., & Schmitt, O. (2024). Modeling and simulation for prediction of multiple sclerosis progression: A review and perspective. Computers in Biology and Medicine, 108416.
2. Izhikevich, E. M. (2003). Simple model of spiking neurons. IEEE Transactions on neural networks, 14(6), 1569-1572.
3. Kannan, V., Kiani, N. A., Piehl, F., & Tegner, J. (2017). A minimal unified model of disease trajectories captures hallmarks of multiple sclerosis. Mathematical Biosciences, 289, 1-8.
4. https://doi.org/10.1371/journal.pcbi.1010507

Speakers

Vishnu Prathapan

Peter Eipert

Markus Kipp

Revathi Appali

Oliver Schmitt

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P246: A Computational Pipeline for Simulating Mouse Visual Cortex Microcircuits with Spiking Neural Networks

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P246 A Computational Pipeline for Simulating Mouse Visual Cortex Microcircuits with Spiking Neural Networks

Margherita Premi*1, Carlo Andrea Sartori1, Giancarlo Ferrigno1, Alessandra Pedrocchi1, Fiorenzo Artoni1, Alberto Antonietti1
1NeuroEngineering and Medical Robotics Laboratory - Department of Electronics, Information and Bioengineering - Politecnico di Milano, Milan, Italy
*E-mail: margherita.premi@polimi.it

Introduction
To integrate in vitro methodologies with in silico techniques, to investigate brain development and neural circuit interactions, we are preparing a computational pipeline to recreate Brain-on-Chip [1] systems with spiking neural networks. We leveraged the MICrONs dataset [2], which provides detailed reconstructions of neurons and astrocytes, with their connections, in a cubic millimeter of mouse visual cortex. The dataset presents significant challenges for computational modeling, particularly regarding quality and quantity of the automatically identified synapses. In this work, we establish a pipeline for transforming raw data into functional spiking neural networks that accurately represent cortical microcircuits.

Methods
The MICrONs dataset showed critical limitations: insufficient synapses and incorrect morphological attributions.
Two solutions were implemented:

● Synapse enhancement through cloning, generating a cluster of synapses placed in a sphere centered on the original synapse. The new synapses are validated through layer densities analyses [3].
● Improve synapse attribution using proofread astrocytes to establish connectivity patterns for non-proofread cells. For neurons, templates from proofread synapses will serve as models for non-proofread neurons.

The framework incorporates layer-specific connectivity with bidirectional astrocyte-neuron interactions. Comparisons were made with networks having the same neurons but different connectivity [4].

Results
Our synapse enhancement method generated clusters of 10 synapses placed in spheres with 10 μm radius, centered on original synapses. This successfully increased the overall synapses count while maintaining layer-specific patterns. A geometric approach was developed that defines minimum ellipsoidal domains containing all synapses belonging to each proofread astrocyte [5]. These ellipsoid representations served as spatial patterns for non-proofread astrocytes. For neurons, template-based attribution from proofread synapses increased the accuracy of connection identification.
Layer-specific connectivity analysis demonstrated that our reconstructed network successfully preserved the characteristic connection patterns across cortical layers (Fig1).
Discussion
This work addresses the identified limitations in using the MICrONs dataset. The developed methods correct connectivity data, enabling more accurate modeling of cortical microcircuits. The approach preserves connections and layer-specific organization unique to the MICrONs dataset. This network is then imported and simulated as a spiking neural model to generate biologically realistic activity. This framework also allows testing alternative network architectures (e.g., random, small-world, etc) compared to the accurate structural connectivity. Future work will refine astrocyte-neuron interaction models. These methodologies could then be applied to BoC experimental data, further validating the computational approaches.

Figure 1. Fig. 1: A. Enhanced synaptic density distribution across cortical depth. B. Astrocyte influence zones represented as ellipsoidal regions, each containing associated synapses. C. Functional connectivity diagram of the reconstructed microcircuit showing layer-specific connections and bidirectional signaling with astrocytes.
Acknowledgements
This work is part of the Extended Partnership "A multiscale integrated approach to the nervous system in health and disease" (MNESYS), funded by the European Union - Next Generation EU under the National Recovery and Resilience Plan, Mission 4, Component 2, Investment 1.4, Project PE00000006, CUP E63C22002170007, Spoke 3 "Neuronal Homeostasis and brain-environment interaction".
References
1. https://doi.org/10.1063/5.0121476
2. https://doi.org/10.1101/2021.07.28.454025
3. https://doi.org/10.1523/JNEUROSCI.0090-23.2023
4. https://doi.org/10.1101/2024.11.18.624135
5. https://github.com/rmsandu/Ellipsoid-Fit

Speakers

Alberto Antonietti

Margherita Premi

Alessandra Pedrocchi

Carlo Andrea Sartori

Giancarlo Ferrigno

Fiorenzo Artoni

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P247: Basic organization of spinal locomotor network derived from hindlimb design and locomotor demands

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P247 Basic organization of spinal locomotor network derived from hindlimb design and locomotor demands

Boris I. Prilutsky*1, S. Mohammadali Rahmati#1, Sergey N. Markin2, Natalia A. Shevtsova2, Alain Frigon3, Ilya A. Rybak2, Alexander N. Klishko#1

1School of Biological Sciences, Georgia Institute of Technology, Atlanta, GA, USA
2Department of Neurobiology and Anatomy, Drexel University, Philadelphia, PA, USA
3Department of Pharmacology-Physiology, University de Sherbrooke, Sherbrooke, QC, Canada

*Email: boris.priutsky@ap.gatech.edu
#These authors contributed equally to this work

Introduction

One of the core principles of sensorimotor physiology is that the musculoskeletal system and its neural control have coevolved to satisfy behavioural demands. Therefore, it may be possible to derive the organization of the neural control of a motor behaviour (e.g., locomotion) from its mechanical demands and properties of the musculoskeletal system. The goals of this study were to (1) determine activity patterns of cat hindlimb muscles from locomotor demands of walking, (2) determine muscle synergies from the predicted and recorded muscle activity patterns and (3) propose a spinal locomotor network organization based on the derived muscle synergies.

Methods
We defined locomotor demands as patterns of resultant moments of force at hindlimb joints generating walking kinematics. To determine the locomotor demands, we computed the resultant muscle moments (using motion capture and methods of inverse dynamics) and muscle activations producing the moments and minimizing muscle fatigue using optimization. We then derived muscle synergies using the non-negative matrix factorization from the computed and recorded activities. We constructed a rhythm generation and pattern formation network of a spinal central pattern generator (CPG) from the derived muscle synergies and incorporated it into our neuromechanical model of spinal hindlimb locomotion.

Results
Locomotor activity patterns of hindlimb muscles obtained from hindlimb musculoskeletal properties and locomotor demands demonstrated a close agreement with the recorded activity patterns. Muscle synergies and their activation patterns derived from the predicted and measured hindlimb muscle activations were similar and consisted of two flexor and three extensor synergies. We used the revealed muscle synergies to construct a spinal CPG and incorporated it into a neuromechanical model of cat hindlimb locomotion. Computer simulations of locomotion demonstrated realistic locomotor mechanics and activity patterns.

Discussion
We demonstrated that hindlimb musculoskeletal properties and locomotor demands (desired resultant joint moments and minimization of muscle fatigue) can predict hindlimb muscle activation patterns, muscle synergies and a general organization of the CPG. The predicted and recorded muscle activations had the following features: (i) reciprocal activation of antagonists, (ii) concurrent activation of agonists and (iii) dependence of activity of two-joint muscles on functional demands. These muscle activation features are typical for many motor reflexes, automatic and highly skilled motor behaviours and suggest that all these behaviours minimize muscle fatigue and have a common organization of spinal circuitries.

Acknowledgements
This work was supported by US National Institutes of Health grants HD032571 and NS110550.
References
No references

Speakers

Boris Prilutsky

Alexander Klishko

Mohammadali Rahmati

Sergey Markin

Natalia Shevtsova

Alain Frigon

Ilya Rybak

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P248: The Neuron as a Self-Supervised Rectified Spectral Unit (ReSU)

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P248 The Neuron as a Self-Supervised Rectified Spectral Unit (ReSU)

Shanshan Qin*1, Joshua Pughe-Sanford1, Alex Genkin1,Pembe Gizem Özdil2, Philip Greengard1, Dmitri B. Chklovskii*1,3

1Center for Computational Neuroscience, Flatiron Institute, New York City, United States
2EDRS Doctoral Program, Swiss Federal Institute of Technology Lausanne(EPFL), Lausanne, Switzerland
3Neuroscience Institute, New York University Grossman School of Medicine, New York City, United States

*Email: qinss.pku@gmail.com; mitya@flatironinstitute.org

Introduction
Advances in synapse-level connectomics [1, 2, 3] and neuronal population activity imaging [4] necessitate neuronal models capable of integrating effectively with these rich datasets. Ideally, these models should offer greater biological realism and interpretability than rectified linear unit (ReLU) networks trained via error backpropagation [5], while avoiding the complexity and parameter intensity of detailed biophysical models[6]. Here, we propose a self-supervised multi-layer neuronal network employing identical learning rules across layers, progressively capturing more complex and abstract features, similar to the Drosophila visual system for which both neuronal responses and connectomics data are available.

Methods
We introduce the Rectified Spectral Unit (ReSU), a neuron model that rectifies projections of its input onto a singular vector of the whitened covariance matrix between past and future inputs. Representing singular vectors corresponding to the largest singular values in each layer effectively maximizes predictive informa- tion [7, 8]. We construct a two-layer ReSU network trained self-supervisedly on translating natural scenes. Inspired by the Drosophila visual system, each first-layer neuron receives input exclusively from one pixel [1], while second-layer neurons integrate inputs potentially from the entire first-layer population. Post-training, we compare the network’s neuronal responses and synaptic weights with empirical results.
Results
First-layer ReSU neurons learned temporal filters closely matching responses observed in Drosophila visual neurons: specifically, the first singular vector matched the linear filter of the L3 neuron, while the second singular vector corresponded to linear filters of L1 (ON) and L2 (OFF) neurons [9]. Additionally, these learned filters adapted their shapes according to signal-to-noise ratios, consistent with experimental find- ings [10]. Second-layer ReSUs aligned with the second singular vector developed motion-selective responses analogous to Drosophila T4 cells [11], and the synaptic weights learned by these neurons closely resembled those documented in T4 connectomic data [2] (Fig.1).
Discussion
ReSU networks exhibit significant advantages, including simplicity, robustness, interpretability, and biolog- ical plausibility. Rectification within ReSUs functions as a form of dynamic clustering, enabling transitions between distinct linear dynamical regimes. Our findings indicate that self-supervised multi-layer ReSU net- works trained on natural scenes faithfully reproduce critical aspects of biological sensory processing. Conse- quently, our model provides a promising foundation for large-scale, interpretable simulations of hierarchical sensory processing in biological brains.

Figure 1. (a)Neurons learn to predict future input and output the (rectified) latent variable. (b)The fly ON motion detection pathway. (c)Responses of neurons to stepped luminance stimulus. (d)T4 response to a moving grating. (e)Temporal filters adaptation to the input SNR. (f)Spatial filter obtained by SVD of L1-L3 output approximates the weights of synapses impinging onto T4a in Drosophila (b).
Acknowledgements
We thank Charles Epstein, Anirvan M. Sengupta and Jason Moore for helpful discussion.
References
[1]https://doi.org/10.1016/j.cub.2013.12.012
[2]https://doi.org/10.7554/eLife.24394
[3]https://doi.org/10.1016/j.cub.2023.09.021
[4]https://doi.org/10.7554/eLife.38173
[5]https://doi.org/10.1038/s41586-024-07939-3
[6]https://doi.org/10.1371/journal.pcbi.1006240
[7]https://doi.org/10.1109/ISIT.2006.261867
[8]Chechik, D., Globerson, A., Tishby, N., & Weiss,Y.(2005).Information Bottleneck for Gaussian Variables. Journal of Machine Learning Research, 6, 165–188
[9]https://doi.org/10.7554/eLife.74937
[10]https://doi.org/10.1098/rspb.1982.0085
[11]https://doi.org/10.1146/annurev-neuro-080422-111929

Speakers

Shanshan Qin

Joshua Pughe-Sanford

Alexander Genkin

Philip Greengard

Dmitri Chklovskii

Pembe Gizem Özdil

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P249: Brain-wide calcium imaging in zebrafish and generative network modelling reveal cell-level functional network properties of seizure susceptibility

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P249 Brain-wide calcium imaging in zebrafish and generative network modelling reveal cell-level functional network properties of seizure susceptibility

Wei Qin*1, Jessica Beevis2, Maya Wilde2, Sarah Stednitz1, Josh Arnold2, Itia Favre-Bulle2, Ellen Hoffman3, Ethan K. Scott1

1 Department of Anatomy and Physiology, University of Melbourne, VIC, Australia
2 Queensland Brain Institute, University of Queensland, QLD, Australia
3 Department of Neuroscience, Yale School of Medicine, Yale University, New Haven, CT, USA

*Email: wei.qin@unimelb.edu.au

Introduction

Epilepsy causes recurrent seizures, but the exact mechanisms are still unclear. Traditional methods using data from primates or rodents struggle to resolve individual cell activity while tracking whole-network dynamics. Capturing the interactions of individual neurons within brain-wide networks could greatly enhance our understanding. Zebrafish, which share genetic and physiological similarities with humans, can exhibit seizure-like behaviors when exposed to drugs like PTZ, which blocks inhibitory GABAergic signaling and induces hyperexcitability [2]. Zebrafish and calcium imaging enable simultaneous in-vivo recording of neuronal activity across the brain at cellular resolution, offering a valuable approach to studying epilepsy [1].

Methods
In-vivo light-sheet calcium imaging was used to capture brain-wide cellular-resolution Calcium fluorescent data from wildtype andscn1lab(a gene implicated in Dravet Syndrome) mutant zebrafish larvae [3]. We conducted this under both baseline and PTZ conditions. Through network analyses, we statistically quantified differences in network topology and dynamics between the two genotypes. We focused on the network of active neuronal cells involved in ictogenesis at microscopic and macroscopic scales. Additionally, we developed a Generative Network Model [4] (GNM, Fig. A, Eq. 1) to explain the wiring principles governing both genotypes and the impact of thescn1labmutation on the brain-wide functional network.
Results
Our study reveals significant changes in brain network connectivity, showing thatscn1labmutations impact brain structure and function. The GNM at the cellular level explains the wiring principles governing the development of both genotypes (Fig. B) and the effects of PTZ on the brain-wide network. The model predicts genotypes and seizure severities for each fish before any seizure activities. This novel model also highlights brain regions associated with genotype differences (Fig. C, D), seizure severity, and overall network excitability. Combining experimental data and mathematical modeling, our approach offers a novel perspective on epileptogenesis mechanisms at a depth and resolution that traditional studies cannot achieve.
Discussion
Our study shows thatscn1lab-/-zebrafish larvae have significant brain morphology changes and increased PTZ-induced seizure susceptibility. Their network architecture mirrors PTZ-treated networks' wiring principles. Brain-wide, cellular-resolution activity data revealed notable alterations in baseline functional wiring, and PTZ administration affected network properties differently inscn1lab-/-and WT larvae, highlighting divergent neural responses. The GNM elucidated specific brain regions where the habenula, pallium, and cerebellum in Dravet Syndrome shows how multiple brain regions are affected, with the habenula influencing seizure initiation and the cerebellum regulating excitatory-inhibitory balance.

Figure 1. A. Generative network modelling (GNM) simulates wiring principles, evaluated by KS similarity. B. The model accurately classifies and predicts genotypes without relying on phenotypes. C. It assesses the contribution of each region to correct classification at each PTZ stage. D. The pallium and habenula are identified as the main contributors to the classification.
Acknowledgements
The authors would like to thank the UQBR aquatics team for maintenance of fish stocks. This project is supported by NHMRC, ARC, Simons Foundation and NIH (US).
References1. https://doi.org/10.1007/978-94-007-2888-2_40 2. https://doi.org/10.1371/journal.pone.0054166 3. https://doi.org/10.1093/braincomms/fcae135 4. Hills, T. T. (2024). Generative Network Models and Network Evolution. In: Behavioral Network Science: Language, Mind, and Society (pp. 46-60). Cambridge University Press.

Speakers

Wei Qin

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P250: A computational study of the influence of circadian adaptive mechanisms on signal processing in retinal circuits

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P250 A computational study of the influence of circadian adaptive mechanisms on signal processing in retinal circuits

Laetitia Raison-Aubry*1, Nange Jin2, Christophe P. Ribelayga2, Laure Buhry1

1Université de Lorraine, CNRS, LORIA, F-54000 Nancy, France
2Vision Sciences, University of Houston, Houston, Texas, United-Stats

*Email: laetitia.raison-aubry@loria.fr

Introduction

Rod-mediated signals reach retinal ganglion cells (RGCs) via three major pathways with distinct sensitivities and operating ranges [1,2,3]. These pathways interact with the cone pathway to ensure seamless processing over >9 log units of light intensity [1]. Gap junctions (GJs) between rod and cone terminals, the entry point of the secondary rod pathway (SRP), exhibit circadian plasticity--stronger at night--directly modulating rod signal flow into cones, and thereby SRP influence on retinal output [4,5]. However, experimentally isolating this effect is challenging due to the non-specificity of pharmacological interventions. Biophysical modeling provides a precise and reversible alternative to selectively manipulate rod/cone coupling while preserving other synaptic conductances. Using a recent mathematical model of a retinal microcircuit [6], we investigate how circadian modulation shapes rod and cone signal integration.
Methods
Our simulated network consists of ~40,000 retinal cells presynaptic to a single transient OFF alpha (tOFF a) RGC [6], arranged on a circular grid approximating the RGC’s receptive field [7] and interconnected with >100,000 synaptic connections, including chemical and electrical synapses. Each retinal cell type is implemented using conductance-based models that follow the Hodgkin-Huxley formalism. Light-induced photocurrent waveforms, whose amplitude and kinetics vary nonlinearly with stimulus intensity [8], serve as input stimuli [6].
Measurements of transjunctional conductance between adjacent mouse rod/cone pairs reveal dynamic changes, ranging over 1000 pS [4,9]. To simulate circadian modulation of rod/cone coupling, we define three states for the GJ channels conductance: uncoupled (0 pS), resting/dark-adapted (300 pS), and maximally coupled (1,200 pS), in line with experimental data [4,9]. Simulations are conducted using Brian 2 [10].
Results
To evaluate the impact of circadian adaptation on retinal signal processing and RGC light responses, we compare normalized intensity-response profiles of the tOFF aRGC across rod/cone coupling states. Stimulus intensity spans the activation threshold of the primary (0.01 R*/rod/s) to the tertiary (60 R*/rod/s) rod pathways [3]. We find that, relative to the SRP resting dark-adapted range (1-60 R*/rod/s) [3], inhibiting rod/cone coupling lowers the sensitivity threshold by ~0.5 log unit, while increasing coupling shifts the tOFF aRGC activation threshold ~1 log unit to the right.
Discussion
Our results support a circadian shift in the threshold and relative contribution of the SRP to the retinal output. This computational approach circumvents experimental limitations, allowing precise investigation of rod/cone coupling modulation. By clarifying mechanistic links between circadian modulation and retinal sensitivity, we demonstrate that our model can be used as a theoretical framework to reconcile previous experimental inconsistencies [5].

Acknowledgements
Research in the Ribelayga lab is supported by National Institutes of Health Grants EY032508, EY029408, and MH127343, National Institutes of Health Vision Core Grant P30EY007551, and The Foundation for Education and Research in Vision (FERV).
References
● https://doi.org/http://dx.doi.org/10.1016/S1350-9462(00)00031-8
● https://doi.org/https://doi.org/10.1146/annurev.physiol.67.031103.151256
● https://doi.org/https://doi.org/10.1126/sciadv.aba7232
● https://doi.org/https://doi.org/10.1126/sciadv.abm4491
● https://doi.org/10.1016/j.preteyeres.2022.101119
● https://doi.org/https://doi.org/10.1109/NER52421.2023.10123863
● https://doi.org/https://doi.org/10.1371/journal.pone.0180091
● https://doi.org/10.1113/jphysiol.2014.284919
● https://doi.org/https://doi.org/10.7554/eLife.73039
● https://doi.org/https://doi.org/10.7554/eLife.47314

Speakers

Laure Buhry

Laetitia Raison-Aubry

Nange Jin

Christophe P. Ribelayga

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P251: Second-order neural field models capture the power spectrum and other nonlinear features of EEG signals in an interval timing task

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P251 Second-order neural field models capture the power spectrum and other nonlinear features of EEG signals in an interval timing task

IanD.Ramsey1andRodicaCurtu1,2,*

1DepartmentofMathematics,UniversityofIowa,IowaCity,USA
2IowaNeuroscienceInstitute,UniversityofIowa,IowaCity,USA

*Email:rodica-curtu@uiowa.edu

Introduction
Neuralfieldmodelsofferapowerfulframeworkforunderstandinglarge-scaleneuronaldynamics by encoding the underlying spatiotemporal processes as a system of integrodifferential equations. While early approaches modeled mean membrane potentials with a single quantity, modern methods [1] distinguish between postsynaptic and somatic potentials to provide a more nuanced description of synaptic interactions and their temporal dynamics. For this work, we consider the second-order neural field model (2ndNFM) introduced by Liley et. al. [2] and investigate how model parameters, governing both local activity and long-range connections, affect thetheta-band and alpha-bandpower of multi-leadEEGsignals as reported by [3].

Methods
We propose a novel method for parameter estimation, utilizing recent developments in the characterization of nonlinear stochastic oscillators [4]. We implement the method to study~4Hz rhythms (2-5 Hz band) of EEG recordings that were found to correlate with cognition inParkinson’s disease (PD) [3]. We extract relevant features (e.g., the Q-function; see [4]) from the EEG data of PD patients and of healthy subjects performing an interval timing task [3], according to the algorithm proposed by [5]. We analyze these nonlinear dynamical features for significant differences between the groups, then perform parameter estimation andextendedKalmanfilteranalysis in the 2ndNFM to obtain a model that captures their characteristics.

Results
We extended the results in [3] by analyzing the EEG signals recorded at the central leads C1 to C6. We found relevant changes in the 2-5 Hz frequency band activity for control and PD groups, like previous reports at Cz. Next, we parametrized the 2ndNFM to capture theattenuated2-5 HzrhythmsseeninPD patients, focusing on the functional coupling between a pair of leads placed on theleft (C3) and right (C4) brain hemispheres. We projected the dynamics of each 10-dimensionalsystem of differential equations perEEGchannel in 2ndNFM on a single variable via Q-function analysis [4, 5]. These projections were used for the model parameter estimation. The resulting 2ndNFM accurately fitted thepowerspectrumoftheEEGsignals at C3 and C4.
Discussion
To test the validity of 2ndNFMs for EEGs in an interval timing task, we performed parameter estimation using recordings at two central leads C3, C4. We also measured the performance of other methods [6] that assumed linearization of 2ndNFMs. We found them to fail to accurately fit the power spectrum of EEG signals due to nonlinear distortions. From our Kalman filter analysis, we detected anomalies in the subcortical and long-range inputs to the linear model that are inconsistent with previous assumptions of statistical independence. The nonlinear 2ndNFM parameterized on data-driven features guarantees an accurate fit for the power spectrum of EEG signals and could generate theoretical predictions.

Acknowledgements
This work was funded by The Stanley-UI Foundation Support Organization (R.Curtu) and the Erwin and Peggy Kleinfeld Endowment (I.Ramsey).
References
1. Cook, B.et al.(2022). Neural field models: a mathematical overview and unifying framework.Math. Neuro. and Appl., 2(2):1-67.
2. Liley, D.et al.(2002) A spatially continuous mean field theory of electrocortical activity.Network: Computation in Neural Syst., 13:67-113.
3. Singh, A.et al.(2021)https://doi.org/10.1038/s41531-021-00158-x
4. Perez-Cervera, A.et al.(2023)https://doi.org/10.1073/pnas.2303222120
5. Melland, P., & Curtu, R (2023)https://doi.org/10.1523/JNEUROSCI.1531-22.2023
6.Hartoyo, A.,et al.(2019)https://doi.org/10.1371/journal.pcbi.1006694

Speakers

Ian Ramsey

Rodica Curtu

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P252: Developments around MOOSE for modeling in Systems Biology and Neuroscience

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P252 Developments around MOOSE for modeling in Systems Biology and Neuroscience

Subhasis Ray*1,2, G.V. Harsharani3,4, Anal Kumar3, Ashish Anshuman3, Parita Mehta3, Jayesh Poojari3, Deepa SM3, and Upinder S. Bhalla3,4

1CHINTA, TCG CREST, Kolkata, India
2IAI, TCG CREST, Kolkata, India
3NCBS-TIFR, Bangalore, India
4Centre for Brain and Mind, NCBS-TIFR, Bangalore, India
*Email: ray.subhasis@gmail.com

Introduction

Public databases for neuroscience, including those for connectomes, cell morphologies, and electrophysiological recordings, are accelerating data-driven neuroscience. Tools supporting such databases and standard formats for model and data exchange are critical for maximizing the utility of these resources. MOOSE, the Multiscale Object Oriented Simulation Environment [1], is a stable software for computational modeling and simulation in Systems Biology and Neuroscience. It emphasizes models that span molecular and electrical signaling from synapses to networks. As MOOSE-development emphasized standards and interoperability early on, it is well placed to facilitate the development of biological neural models utilising public model- and data-repositories.
Methods
The core of MOOSE is written in C++ for speed, while its Python API allows integration with the Python ecosystem. Extensive documentation is supplemented with a wide range of tutorials using Python graphics and browser-based 3-D graphics. We use existing Python modules for various model and data description formats to support them in MOOSE, and web frameworks to utilize public APIs of the neuroscience databases. We actively conduct outreach activities and user-research to enhance the user experience and documentation of MOOSE, and workshops for training students and researchers on modeling and simulation in Systems Biology and Computational Neuroscience.
Results
MOOSE covers multiple scales of modeling, from chemical reactions and signaling pathways to large biological neural networks. Currently it supports standard formats like SBML and NeuroML for model description, SWC for morphology, and NSDF for simulated data. It includes Python tools to easily create multiscale models from a library of model components. We are also developing clients for accessing public repositories of model and data, enabling users to seamlessly integrate model components from such sources into their composite models.
Discussion
A major goal of MOOSE is to make biological modeling accessible to students and researchers from diverse backgrounds. Users can seamlessly incorporate published and curated models in their simulation experiments using software tools developed around MOOSE. The new developments in the MOOSE ecosystem will help accelerate data-driven research in Systems Biology and Neuroscience.

Acknowledgements
We thank the Kavli Foundation, and DBT and DST of the Govt. of India, for supporting MOOSE development. NCBS/TIFR receives support from the Department of Atomic Energy, Government of India, under Project Identification No. RTI 4006.

References
● https://doi.org/10.3389/neuro.11.006.2008

Speakers

Upinder Bhalla

Anal Kumar

Subhasis Ray

Gv Harsharani

Ashish Anshuman

Speakers

Thomas Richner

Nicholas Gregg

Raunak Singh

Keith Starnes

Dora Hermes

Jamie Van Gompel

Gregory Worrell

Brian Lundstrom

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P256: Structural and evolutionary insights of the neuropeptide connectome of Caenorhabditis species

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P256 Structural and evolutionary insights of the neuropeptide connectome of Caenorhabditis species

Lidia Ripoll-Sánchez*1,2,Itai A. Toker4,Oliver Hobert4,Isabel Beets3,Petra E. Vértes1,2,5,William R. Schafer1,3,5

1MRC Laboratory of Molecular Biology, Cambridge, UK
2Department of Psychiatry, Cambridge University, Cambridge, UK
3Department of Biology, KU Leuven, Leuven, Belgium
4Department of Biological Sciences, Howard Hughes Medical Institute, Columbia University, New York, NY, USA
5co-senior authors

*Email: lsanchez@mrc-lmb.cam.ac.uk

Introduction

Neuropeptides modulate synaptically wired neuronal circuits. This modulation is critical to nervous system function, yet little is known about the structure and function of extrasynaptic signalling networks at a whole-organism level and how that is maintained over evolution.

Methods

To this end, we used single neuron gene expression [1] and deorphanisation data for neuropeptide-activated G-protein coupled receptors [2] to generate a connectome of 92 neuropeptide signalling networks inC. elegans[3].This network defined a connection when the sending neuron expressed a neuropeptide, the receiving neuron expressed the cognate receptor, and both neurons extended overlapping processes.We then used graph theory and machine learning methods to characterise its structural features.

Results

Our analysis on the connectivity pattern revealed a mesoscale structure for the core of the network, splitting it in three groups of neurons that act as highly controlled functional hubs. Notably inside these hubs, we identified a group of neurons that seem to be morphologically and biochemically adapted for neuropeptidergic communication. Furthermore, the co-expression pattern identified autocrine neuropeptidergic connections that may modulate locomotion control and evolutionary conserved intracellular neuropeptide signalling networks that could act as homeostatic regulators of the neuropeptidergic network. This network has a higher connection density than the synaptic and gap junction ones, connecting non-synaptically connected neurons [3].

Discussion

These findings challenge the idea that neuronal communication is primarily synaptic, revealing a dense, decentralised neuropeptide network with functional and structural roles. Additionally, conserved signalling patterns acrossCaenorhabditisspecies highlight the evolutionary significance of neuropeptide connectivity[4].Weexpect that this newly mapped neuropeptide connectomes, their analysis and the interactive website we developed to explore them (nemamod.org) will serve as a prototype for other animals and provide new insight into the structure of neuromodulatory networks in larger brains.

Acknowledgements
This work was funded by the Howard Hughes Medical Institute and the NIH grants RO1 NS039996 & NIH RO1 NS100547 (to OH); the Medical Research Council grant MC-A023-5PB91 (to WRS); a Medical Research Council PhD fellowship (to LRS); the MQ Transforming Mental Health grant MGF17_24 (to PEV); and a postdoctoral fellowship from the Evelyn Gruss Lipper charitable foundation (to IAT).
References
1.https://doi.org/10.1016/j.cell.2021.06.023
2.https://doi.org/10.1016/j.celrep.2023.113058
3.https://doi.org/10.1016/j.neuron.2023.09.043

4.https://doi.org/10.1101/2024.11.23.624988

Speakers

Lidia Ripoll-Sánchez

Itai Antoine Toker

Oliver Hobert

Isabel Beets

Petra E. Vértes

William R. Schafer

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P257: Basket cell computational modeling predicts signal filtering of Purkinje cell responses

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P257 Basket cell computational modeling predicts signal filtering of Purkinje cell responses

Martina F. Rizza*1,Stefano Masoli1, Teresa Soda1, Francesca Prestori1, Egidio D’Angelo1,2

1Department of Brain and Behavioural Sciences, University of Pavia, Pavia, Italy
2Digital Neuroscience Center, IRCCS Mondino Foundation, Pavia, Italy

*Email: martinafrancesca.rizza@unipv.it

Introduction
Cerebellar basket cells (BC), located in the bottom 1/3 of the molecular layer (ML), play an important role in controlling the activity of Purkinje cells (PC) via inhibitory synaptic transmission. BCs receive excitatory synaptic inputs from parallel fibers (pf) and transmit inhibitory synaptic inputs to BCs and PCs. We reconstructed a multi-compartmental biophysically realistic BC model in Python-NEURON to investigate BC intrinsic and synaptic electrophysiological properties and their impact on PC model responses [1,2,3]. The SC model [4] was included to reconstruct a ML microcircuit. Simulationspredicted that BC and SC operate in tandem, setting the frequency band of PC transmission through the regulation of the PC frequency/response curve.

Methods
Starting from morphological reconstructions taken from cerebellar tissue and patch-clamp recordings, we implemented conductance-basedmulti-compartmental models of BCwith Python 3 and NEURON 8.2[5].The model maximum ionic conductances were tuned to match the firing pattern revealed by whole-cell patch-clamp recordings from mice cerebellar slices.Mouse SC[4] and BC models were connected with a multi-compartmental mouse PC model[1,2,3] to test their impact when stimulated by excitatory synaptic inputs.Simulations were performed on an AMD Threadripper 7980X 64 cores using fixed time step (0.025ms) and temperature set to 32°.

Results
Simulations reproduced whole-cell patch-clamp experimental results, showing autorhythmic activity, an almost linear I/O relationship to positive current injections, pauses generated after positive current injections,sag after the negative current injections,AMPA and NMDA receptors-mediated excitatory postsynaptic responses following pf inputs.SC and BCsimulations showed thefiltering properties on PC activity, highlighting thatBCs modulate low-frequency PC discharges through somatic GABAergic synapses, while SCs act on high-frequency responses through dendritic GABAergic synapses.

Discussion
BC modeling reproduced the cellular intrinsic excitability and the synaptic activity, investigating thefrequency‑dependent short‑term dynamics at pf-BC synapses and the frequency-dependenceof BC input–output gain functions. Simulations predicted BC and SC filtering of PC responses, showing thatthe intensity and bandwidth of ML filtering is modulated by the number of active synapses between pfs-SCs-PCs and pfs-BCs-PCs. SCs and BCs emerge as critical elements controlling cerebellar processing in time and frequency domains. Tuning of transmission bandwidth and delay through specific membrane and synaptic mechanisms contributes to explain the role of SCs and BCs in motor learning and control.

Acknowledgements
This project/research received funding from the European Union’s Horizon Europe Programme under the Specific Grant Agreement No. 101147319 (EBRAINS 2.0 Project) and the European Union’s Horizon 2020 Framework Programme for Research and Innovation under the Framework Partnership Agreement No. 650003 (HBP FPA).
References
1.https://doi.org/10.3389/fncel.2015.00047
2.https://doi.org/10.1038/s42003-023-05689-y
3.https://doi.org/10.3389/fncel.2017.00278
4.https://doi.org/10.1038/s41598-021-83209-w
5.https://doi: 10.3389/neuro.11.001.2009

Speakers

Egidio D'Angelo

Stefano Masoli

Martina Francesca Rizza

Teresa Soda

Francesca Prestori

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P258: Reconstruction and simulation of the mouse cerebellar declive: an atlas-based approach

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P258 Reconstruction and simulation of the mouse cerebellar declive: an atlas-based approach

Dimitri Rodarie1*, Dianela Osorio1, Egidio D’Angelo1,2, Claudia Casellato1

1Department of Brain and Behavioral Sciences, University of Pavia, Pavia, Italy
2Brain Connectivity Center IRCCS Mondino Foundation, Pavia, Italy
* Email:dimitri.rodarie@unipv.it

Introduction

We aim to reconstruct and simulate atlas-mappedmousecerebellar regions, capturing the relationship between structure, dynamics, and function. Numerous experiments on rodents and humans show that the declive region (lobule VI) plays a relevant role in many functions including motor, cognitive, emotional, and social tasks [1].
We present here a pipeline to reconstruct the mouse declive, based on the Blue Brain Cell Atlas (BBCA) model [2] and the Brain Scaffold Builder (BSB) tool [3]. With this pipeline, we could estimate the specific densities of each cell type. With the BSB we placed, oriented, and connected the neurons. The output of this pipeline is a circuit that can be simulated and validated against experimental findings.

Methods
We built a 3D model of the mouse declive (Fig. 1), based on the BBCA pipeline (Fig. 1DE), which we extended with the Purkinje layer at the boundary between granular and molecular layers (Fig. 1A). We placed cells based on the atlas and regional densities [4,5] and proposed a new strategy to place Purkinje layer cells based on linear density [6] (Fig. 1F). To connect the cells, we computed the orientations and depth [2] of each morphology (Fig. 1BC). These fields are used to bend the cells’ neurites following the declive curvature (Figure 1G). We applied voxel intersection on these bent cells with synaptic in- and out-degree ratios [3]. Finally, we assigned point-neuron electrical parameters to each cell and connection [7].
Results
We combined the workflows of the BBCA and the BSB into a single pipeline. This includes tools to align experimental data into an atlas, to reconstruct and to simulate cerebellar circuits. This allowed us to produce the most detailed model of the mouse declive.
We obtained new densities for each cell type of the cerebellum. Our model shows cell composition differences between cerebellar regions. We also estimated the impact of the declive shape on its local connectivity, by comparing different sub-part of the region with respect to a cubic canonical circuit.
Finally, we simulated our circuit using the BSB interfacing with the NEST simulator [8] in resting state and created a paradigm to reproduce fear conditioning experiments on mice.
Discussion
By combining the two pipelines to reconstruct our circuit, we are now able to leverage atlas data to estimate the spatial cellular composition in the cerebellum. The atlas registration will also facilitate the embedding of our model into larger brain circuits [9].
We also found that the cerebellum's highly parceled layers, its curved shape and its position within the mouse atlas make our model very sensitive to artifacts in the data (Figure 1DE). The model will be refined as more data become available.
We plan to reconstruct different sub regions of the cerebellar cortex to compare their structure and function. Our future work will also involve mapping the different types of Purkinje neurons based on the “zebrin stripes” [10].

Figure 1. Fig 1: Reconstruction pipeline. A. Annotations shown in colors over the Nissl volume. B. Orientation field showing the local axons’ main axis. Colors represent the vectors’ norm. C. Distance to the outside border, following the orientation field. D. E. Neuron and inhibitory neuron density. F. Neuron positions displayed over annotations. G. Scaled and bent Purkinje morphologies over annotations.
Acknowledgements
Funding:
●
European Union's Horizon 2020 research and innovation program - Marie Sklodowska-Curie - grant 956414 Cerebellum and Emotional Networks
●
Virtual Brain Twin Project - European Union's Research and Innovation Program Horizon Europe - grant 101137289
●
National Centre for HPC, Big Data and Quantum Computing - CN00000013 PNRR MUR – M4C2 – Fund 1.4 - decree n. 3138 16 december 2021

References
1.https://doi.org/10.3389/fnsys.2023.1185752
2.https://doi.org/10.1371/journal.pcbi.1010739
3.https://doi.org/10.1038/s42003-022-04213-y
4.https://doi.org/10.1007/s00429-013-0531-9
5.https://doi.org/10.1523/JNEUROSCI.20-05-01837.2000
6.https://doi.org/10.1038/s41593-022-01057-x
7.https://doi.org/10.3389/fncom.2019.00068
8.https://doi.org/10.4249/scholarpedia.1430
9.https://doi.org/10.1523/ENEURO.0111-17.201710.https://doi.org/10.1038/nrn269

Speakers

Claudia Casellato

Egidio D'Angelo

Dimitri Rodarie

Dianela Osorio

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P259: Local multi-gridding for detailed morphology, spines and synapses

P262: Mechanisms of bistability in spinal motoneurons and its regulation

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P262 Mechanisms of bistability in spinal motoneurons and its regulation

Ilya A. Rybak*1, Yaroslav I. Molkov2, Thomas Stell2, Florent Krust3, Frédéric Brocard3
1 Department of Neurobiology and Anatomy, Drexel University College of Medicine,
Philadelphia, PA, USA
2 Department of Mathematics and Statistics and Neuroscience Institute, Georgia State
University, Atlanta, GA, USA
3 Institut de Neurosciences de la Timone, Aix Marseille University, CNRS, Marseille, France

*Email: rybak@drexel.edu

Introduction

Spinal motoneurons represent output elements of spinal circuitry that activate skeletal muscles to produce motor behaviors. Firing behavior of many motoneurons is characterized by bistability allowing them to maintain a self-sustained spiking activity initiated by a brief excitation and stopped by a brief inhibition. Serotonin can induce or amplify bistability, influencing motor behaviors. Biophysical mechanisms of bistability involve nonlinear interactions of specific ionic currents. Experimental studies identified ionic currents linked to bistability [1,2]. Using computational modeling, we simulate motoneuronal bistability and analyze the roles of key ionic currents in its generation and regulation.

Methods
We have developed a conductance-based mathematical model of spinal motoneuron to explore and analyze the role of different ionic currents and their interactions in generation and control of motoneuronal bistability under different conditions. The one-compartmental model includes main spike-generating currents, fast sodium (INaF) and potassium rectifier (IKdr), as well as persistent sodium (INaP), slowly inactivating potassium (IKv1.2aka potassium A,IKA), high-voltage activated calcium (ICaL), Ca2+-activated cation non-specific (ICAN), and Ca2+-dependent potassium (IKCa, associated with SK channels) currents. Additionally, the model incorporates the intracellular Ca2+dynamics including calcium-induced calcium release mechanism (CICR).
Results
Our simulations show that bistability in motoneurons relies onICAN, activated by intracellular Ca2+accumulated byICaLand the CICR mechanism. Two other currents play modulatory roles withINaPaugmenting bistability andIKCaattenuating or abolishing it. The interplay betweenICANandIKCashapes the membrane potential dynamics, producing post activation afterdepolarization (ADP) or afterhyperpolarization (AHP), withIKv1.2modulating the membrane potential dynamics. Under certain conditions (such as an elevated extracellular K+concentration),INaPcan sustain bistability independently ofICAN.
Discussion
Our findings clarify the ionic basis of motoneuron bistability, underscoring its reliance on current interactions and external conditions, and offer insights into motor function and potential therapeutic strategies for motor disorders. Our results suggest that serotonin can induce or increase motoneuron bistability by amplifyingICAN(e.g., via increased intracellular Ca2+concentration due to an increasedICaLor via 5-HT3 receptors), activation ofINaPor suppression ofIKCa(both through 5-HT2 receptors).

Acknowledgements
No
References
● Harris-Warrick, R.M., Pecchi, E., Drouillas, B., Brocard, F., & Bos R. (2024). Effect of size on expression of bistability in mouse spinal motoneurons. Journal of Neurophysiology, 131(4), 577-588.https://doi.org/10.1152/jn.00320.2023.
● Bos, R., Drouillas, B., Bouhadfane, M., Pecchi, E., Trouplin, V., Korogod, S.M., & Brocard, F. (2021) Trpm5 channels encode bistability of spinal motoneurons and ensure motor control of hindlimbs in mice. Nature Communication, 12(1), 6815.https://doi.org/10.1038/s41467-021-27113-x.

Speakers

Ilya Rybak

Yaroslav Molkov

Thomas Stell

Florent Krust

Frédéric Brocard

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P263: Enhancing Neuronal Modeling with a Modified Hodgkin-Huxley Approach for Ion Channel Dynamics

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P263 Enhancing Neuronal Modeling with a Modified Hodgkin-Huxley Approach for Ion Channel Dynamics

Batoul M. Saab*1, Jihad Fahs2, Arij Daou1

1Biomedical Engineering Program, American University of Beirut, Lebanon
2Department of Electrical and Computer Engineering, American University of Beirut

*Email: bms28@mail.aub.edu

Introduction
The development of precise physical models is imperative for comprehending and manipulating system behavior. Neuronal firing models serve as a pivotal exemplar of intricate biological modeling, crucial for unraveling neural functionality across both normal cognitive processes and pathological disease states. Achieving accurate dynamical modeling of neuronal firing necessitates the meticulous fitting of model parameters through data assimilation, utilizing experimentally gathered recordings. This endeavor poses significant theoretical challenges due to two primary factors: (a) neuronal action potentials are the aggregate result of active nonlinear dynamics interconnecting various neuronal compartments, parameterized by a multitude of unknown variables, and (b) the stochastic nature of the noisy environmental stimuli influencing neuronal activity.

Methods
In practice, the fitting of a substantial number of parameters is constrained by the scarcity of observable outputs (recording sites), the complexity of the underlying models, and the time-intensive and expensive nature of conducting experiments under controlled conditions [1]. While neurophysiologists are restricted to a limited range of feasible injection current waveforms, we propose herein to investigate the parameter estimation conundrum of model neurons using diverse quality metrics and processing techniques. Our approach involves optimizing a biophysically realistic model for these neurons [2] using intracellular data obtained via the whole-cell patch-clamp technique from basal-ganglia projecting cortical neurons in brain slices of zebra finches.
Results
We proceed with adopting a different approach than that adopted by Hodgkin and Huxley [3] in their seminal work whereby we model the activation functions directly using Hill functions rather than fitting both opening rate constants by exponential functions. Our approach provides additional flexibility and is biologically interpretable. Furthermore, using this modified model, we conduct exhaustive searches on a large subset of the model parameters and test different functional metrics to check which one(s) generate reliable and realistic fits to the biological data.
Discussion
The long-term benefits of this approach include the capability to examine large-scale dynamic phenomena in insightful manners, enhancing model accuracy and streamlining experimentation time. By refining parameter estimation methods and employing biologically interpretable mathematical representations, we aim to improve our understanding of neuronal firing dynamics and provide a robust framework for future computational neuroscience research.

Acknowledgements
This work was supported by the University Research Board (URB) and the Medical Practice Plan (MPP) grants at the American University of Beirut.
References
Reference 1:https://doi.org/10.48550/arXiv.1609.00832
Reference 2:10.1152/jn.00162.2013
Reference 3:10.1113/jphysiol.1952.sp004764

Speakers

Arij Daou

Batoul Saab

Jihad Fahes

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P264: Paths to depolarization block: modeling neuron dynamics during spreading depolarization events

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P264 Paths to depolarization block: modeling neuron dynamics during spreading depolarization events

Maria Luisa Saggio*1, Damien Depannemaecker1, Roustem Khazipov2,3, Daria Vinokurova2, Azat Nasretdinov2, Viktor Jirsa1, Christophe Bernard1

1 Aix Marseille Univ, INSERM, INS, Inst Neurosci Syst, Marseille, France
2Laboratory of Neurobiology, Institute of Fundamental Medicine and Biology, Kazan Federal University, Kazan, 420008, Russia
3Aix-Marseille University, INMED, INSERM, Marseille, 13273, France

*Email: maria-luisa.saggio@univ-amu.fr

Introduction
Spreading Depolarization (SD) is a pathological state of the brain involved in several brain diseases, including epilepsy and migraine. It consists of a slowly propagating wave of nearly complete depolarization of neurons, classically associated with a depression of cortical activity. Recent findings challenge this homology [1]: during SD events, which only partially propagate from the cortical surface to depth, neuronal activity may be suppressed, unchanged or elevated. In layers invaded by SD, neurons lose their ability to fire entering Depolarization Block (DB) and far from the SD neurons maintain their membrane potential. However, neurons in between unexpectedly displayed patterns of prolonged sustained firing.

Methods
In the present work [2], we build a phenomenological model, incorporating some key features observed during DB in this dataset (current-clamp patch-clamp recordings from L5 pyramidal neurons in the rat somatosensory cortex during evoked SDs), that can predict the new patterns observed. We model the L5 neuron as an excitable system close to a SNIC bifurcation [3], using the normal form of the unfolding of the degenerate Takens-Bogdanov singularity for the fast dynamics [4], a minimal yet dynamically rich dynamical system. The fast subsystem is modulated by the dynamics of two slow variables, implementing homeostatic and non-homeostatic reactions to inputs.
Results
The model’s bifurcation diagram provides a map for neural activity that includes baseline behavior, sustained oscillations, and DB. We identify five qualitatively different scenarios for the transition from healthy activity to DB, through specific sequences of bifurcations. These scenarios encompass and expand on the mechanisms for DB present in the modeling literature, account for the novel patterns observed in our dataset,and allow us to understand them from a unified perspective. Time series in our dataset are consistent with the scenarios, however, the presence of bistability, distinguishing some of the scenarios, cannot be inferred by our analysis. We further use the model to investigate mechanisms for the return to baseline.
Discussion
Understanding how brain circuits enter and exit SD is important to designing strategies aimed at preventing or stopping it. In this work, we use modeling to gain mechanistic insights into the ways a neuron can transition to DB or different patterns of sustained oscillatory activity during SD events, as observed in our dataset. While our work provides a unified perspective to understanding the modeling of DB, ambiguities remain in the data analysis. These ambiguities could be solved by scenario-dependent theoretical predictions, for example for the effect of stimulation, for further experimental testing.

Acknowledgements
Funded by the Russian Science Foundation grant № 24-75- 10054 to AN (https://rscf.ru/en/project/24-75-10054/) and the European Union grant № 101147319 to MS, DD and VJ.
References
[1] Nasretdinov, A., Vinokurova, D., Lemale, C. L., Burkhanova-Zakirova, G., Chernova, K., Makarova, J., ... & Khazipov, R. (2023). Diversity of cortical activity changes beyond depression during spreading depolarizations. Nature Communications, 14(1), 7729.
[2] Saggio et al (In preparation)
[3] Izhikevich, E. M. (2007). Dynamical systems in neuroscience. MIT press.
[4] Dumortier, F., Roussarie, R., & Sotomayor, J. (1991). Generic 3-parameter families of planar vector-fields, unfoldings of saddle, focus and elliptic-singularities with nilpotent linear parts.

Speakers

Viktor Jirsa

Maria Luisa Saggio

Damien Depannemaecker

Roustem Khazipov

Daria Vinokurova

Azat Nasretdinov

Christophe Bernard

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P265: Bifurcations and bursting in the Epileptor

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P265 Bifurcations and bursting in the Epileptor

Maria Luisa Saggio*1, Viktor Jirsa1

1Aix Marseille Univ, INSERM, INS, Inst Neurosci Syst, Marseille, France

*Email: maria-luisa.saggio@univ-amu.fr

Introduction

Large-scale patient-specific models could provide clinicians with an additional tool for the evaluation of the best surgery strategy for drug-resistant epileptic patients, leveraging on the possibility of revealing otherwise hidden complex network and dynamical effects, testing clinical hypotheses, or finding unbiased optimal surgery strategies. One of these frameworks, the Virtual Epileptic Patient (VEP), is currently undergoing validation in a prospective clinical trial involving over 300 patients. It employs the Epileptor [1], a phenomenological mesoscopic model for the most common seizure class, defined in terms of the onset/offset bifurcations marking the transitions between interictal and ictal states are modeled as bifurcations.

Methods
The Epileptor onset/offset bursting class is known in the literature in dynamical systems as square-wave bursting. In this study [2], we utilize insights from a more generic model for square-wave bursting, based on the unfolding of a high codimension singularity, to guide the bifurcation analysis of the Epileptor and gain a deeper understanding of the model and the role played by its parameters. We use analytical methods, numerical continuation of bifurcation curves and model simulations.
Results
We identify a key region in parameter space of topological equivalence between the two models and demonstrate how the Epileptor’s parameters can be modified to produce activities for other seizure classesas predicted by the generic model approach.. Finally, we reveal how the interaction with an additional mechanism for spike-and-wave discharges present in the Epileptor alters the bifurcation structure of the main burster pushing it across a sequence of Supercritical Hopf bifurcations that modulate the oscillatory activity typical of the ictal state.
Discussion
Exploring the full potential of the Epileptor model in terms of bursting dynamics and understanding how to set the parameters to obtain different classes is an important step to (i) enhance our understanding of the model at the core of the VEP framework and (ii) explore the possibility of further personalizing the VEP model. In fact, patients may experience seizures compatible with classes other than square-wave [3]. While the impact of the class on the VEP outcome has not yet been investigated, we know that different classes may exhibit variations in synchronization and propagation properties, warranting further exploration.

Acknowledgements
This research has received funding from EU’s Horizon 2020 Framework Programme for Research and Innovation under the Specific Grant Agreements No. 101147319 (EBRAINS 2.0 Project).
References
[1]Jirsa, V. K., Stacey, W. C., Quilichini, P. P., Ivanov, A. I., & Bernard, C. (2014). On the nature of seizure dynamics.Brain,137(8), 2210-2230.
[2]Saggio, M. L., & Jirsa, V. (2024). Bifurcations and bursting in the Epileptor.PLOS Computational Biology,20(3), e1011903.
[3] Saggio, M. L., Crisp, D., Scott, J. M., Karoly, P., Kuhlmann, L., Nakatani, M., ... & Stacey, W. C. (2020). A taxonomy of seizure dynamotypes.Elife,9, e55632.

Speakers

Viktor Jirsa

Maria Luisa Saggio

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P266: Two distinct bursting mechanisms in cold-sensitive neurons of Drosophila Larva

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P266 Two distinct bursting mechanisms in cold-sensitive neurons of Drosophila Larva

Akira Sakurai, Natalia V. Maksymchuk, Sergiy M. Korogod, Daniel N. Cox,Gennady S. Cymbalyuk*
Neuroscience Institute, Georgia State University, Atlanta, GA 30302-5030, USA

*Email: gcymbalyuk@gmail.com

Introduction
In Drosophila larvae, noxious low temperatures are detected by CIII primary sensory neurons lining the inside of the body wall [1,2]. About half of these neurons respond to rapid temperature drops with transient bursts, producing a clear spike-rate peak that likely signals the rapid change. Previously, we developed a biophysical model, which captured various extracellularly recorded cold-evoked CIII responses [2]. Here, having overcome the challenge posed by the small size of CIII neurons and obtained intracellular recordings, we used the waveforms of bursting to identify two distinct types of bursting generated by these neurons.
Methods
We used electrophysiological intracellular and extracellular recordings and modeling to investigate the mechanisms underlying pattern generation by CIII neurons. We upgraded the model [2], by including dynamics of concentrations of Cl-, Na+, and K+, since Ca2+-activated Cl-current (ICaCl) was implicated in CIII dynamics [3]. We investigated the patterns caused by injected current, a drop in extracellular Cl-, and drop of temperature. We also considered a simplified model with an effective (e) leak current including Cl-currents lumped together with Na+and K+leak currents. We map oscillatory and silent regimes under variation of EeLeakand geLeakand compare the model activity to the experimental data.
Results
At ambient temperatures, CIII neurons exhibited a stationary state around -40 mV and sporadic spikes at 1.0 ± 1.3 Hz (N = 20). In the activity of 90% of sporadically spiking neurons, elliptic bursts with an intra-burst spike frequency of 6.0 ± 1.7 Hz were detected. With a temperature drop from 24°C to 10°C, CIII neurons depolarized and spiked at 2.9 ± 1.5 Hz. In 45% of neurons, square-wave bursts with the intra-burst spike frequency 38.2 ± 19.5 Hz were observed. Similar square-wave bursting and high-frequency spiking were induced by direct depolarizing injected currents. Low-Cl⁻conditions induced transitions between patterns of activity dominated by spiking, fast bursting, or slow bursting.
The model represents waveform properties of the experimentally recorded bursting under variation of the injected current, extracellular Cl-, and temperature. We found large parameter domains of silent and spiking regimes at low and high EeLeak, respectively, and a domain of square-wave bursting in an intermediate range of geLeakas EeLeak. In a certain range of geLeakas EeLeakgrows the model transitions from silence to elliptic bursting and then to spiking. These transitions qualitatively map transitions observed in experimental data.
Conclusion

We identified two distinct types of bursting patterns—elliptic bursting and square-wave bursting in responses of CIII neurons. These findings enhance our understanding of the temperature sensing in insect peripheral sensory neurons, providing insights into how sensory systems respond to environmental stimuli.

Acknowledgements
NIH grant R01NS115209 to DNC and GSC.
References
·1. Turner, H. N., et al. (2016).Current Biology, 26(23): 3116-3128.https://doi.org/10.1016/j.cub.2016.09.038
·2. Maksymchuk, N., et al. (2022).Frontiers in Cellular Neuroscience,16, 831803.https://doi.org/10.3389/fncel.2022.831803

·3. Himmel, N. J., et al., (2023).eLife, 12, e76863.https://doi.org/10.7554/eLife.76863

Speakers

Akira Sakurai

Natalia V. Maksymchuk

Sergiy M. Korogod

Daniel N. Cox

Gennady Cymbalyuk

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P267: Real-time closed-loop perturbation of electrically coupled neurons to characterize sequential dynamics in CPG circuits

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P267 Real-time closed-loop perturbation of electrically coupled neurons to characterize sequential dynamics in CPG circuits

Pablo Sanchez-Martin*¹, Alicia Garrido-Peña¹,Manuel Reyes-Sanchez, Irene Elices¹, Rafael Levi¹, Francisco B Rodriguez¹, Pablo Varona¹
1. Grupo de Neurocomputación Biológica, Departamento de Ingeniería Informática, Escuela Politecnica Superior, Universidad Autónoma de Madrid, Madrid, Spain
*Email: pablo.sanchezm@uam.es

Introduction
Dynamical invariants in the form of robust cycle-by-cycle relationships between intervals that build robust neural sequences have beenobservedrecently in central pattern generatorscircuits (CPGs) [1]. In this study, we analyze the effect of different closed-loop perturbationson electrically coupled neurons that are part of aCPGtodeterminethe associated modulation of sequence interval variability,synchronizationand dynamical invariants.

Methods
This research was performed in the pyloric CPG involving both voltage recordings and current injection in the PD neurons, which are electrically coupled cells in this circuit. Additionally, we recorded extracellularly from the LP neuron to quantify the LPPD delay, an interval that builds a dynamical invariant with the cycle-by-cycle period. We implemented an active electrical compensation procedure [2] on RTXi real-time software, which prevents the recording artifact using a single electrode. Three closed-loop perturbations were delivered on the PD neurons: 1. A Hindmarsh-Rose (HR) model neuron electrically coupled to a PD neuron, thus building a biohybrid circuit. 2. A square pulse current injection during the PD burst. 3. An additional artificial electrical synapse between the two PD neurons.

Results
The electrical coupling with a negative artificial bidirectional synapse did not change the existing invariant relation between the LPPD delay and the period but increased the rhythm variability and increased the Victor-Purpura distance, i.e., reduced the PD synchronization level.The squared pulse perturbation decreased the variability and thus the LPPD delay linear relationship was reduced.The level of synchronization between both PDs was also reduced with the pulse perturbation with respect to the control.The biohybrid circuit built by adding anadditionalelectrical coupling to an artificial HR neuron also reduced the variability but changed the intercept of the linear relationshipi.e.,for the same LPPD delays thePD period was sorter.

Discussion
In this study, we effectively disrupted the dynamics of two electrically coupled neurons with three different perturbations by injecting current into the neurons that modulated the synchronization level.This not onlymodifiedthe dynamics of these neurons but also the whole circuit variability and the associated dynamical invariants.All protocols have been proven effective to study the relationship of electrical coupling and sequential dynamics with the help of real-timeclosed-loopneurotechnologies.

Acknowledgements
Work funded by PID2024-155923NB-I00, CPP2023-010818, PID2023-149669NB-I00 and PID2021-122347NB-I00.
References
[1] I. Elices, R. Levi, D. Arroyo, F. B. Rodriguez, and P. Varona. Robust dynamical invariants in sequential neural activity. Scientific Reports, 9(1):9048, 2019.
[2] R. Brette, Z. Piwkowska, C. Monier, M. Rudolph-Lilith, J. Fournier, M. Levy, and A. Destexhe. High-resolution intracellular recordings using a real-time computational model of the electrode. Neuron, 59(3):379–391, 2008.

Speakers

Alicia Garrido-Peña

Pablo Varona

Francisco B Rodriguez

Pablo Sanchez-Martin

Manuel Reyes-Sanchez

Irene Elices

Rafael Levi

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P268: Modelling Nitric Oxide Diffusion and Plasticity Modulation in Cerebellar Learning

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P268 Modelling Nitric Oxide Diffusion and Plasticity Modulation in Cerebellar Learning

Carlo Andrea Sartori1*, Alessandra Maria Trapani1, Benedetta Gambosi1, Alessandra Pedrocchi1, Alberto Antonietti1

1 Department of Electronics, Information and Bioengineering, Politecnico di Milano, Milano, Italy

* Email:carloandrea.sartori@polimi.it
Introduction

Nitric Oxide (NO) is an important molecule in processes such as synaptic plasticity and memory formation[1]. In the cerebellum, NO is produced by neural NO Synthase expressed in Granule Cells and Molecular Layer Interneurons[2]. NO diffuses freely in tissues beyond synaptic connections, functioning as a volume neurotransmitter. At parallel fiber-Purkinje Cell (pf-PC) synapses[4][5], NO is necessary but not sufficient for both Long Term Potentiation and Depression[6][7]. This study investigates NO role in cerebellar learning mechanisms using a biologically realistic Spiking Neural Network, implementing a NO-dependent plasticity model and testing it with an Eye-Blink Classical Conditioning (EBCC) protocol[8][9].

Methods
We developed the NO Diffusion Simulator (NODS), a Python module modeling NO production and diffusion within a Spiking Neural Network. The model represents the chemical cascade triggered by calcium influx during spikes, leading to NO production[10]. NO diffusion is modeled using the heat diffusion equation and an inactivation term, solved with Green's function[11]. We implemented a NO-dependent supervised Spike-Timing Dependent Plasticity[12]where a termweights synaptic updates based on NO concentration. The model was tested using the EBCC protocol, where the cerebellum learns to associate a Conditioned Stimulus (CS) with an Unconditioned Stimulus (US), generating anticipatory Conditioned Responses (CR) (Fig. 1).
Results
We first validated the equation in NODS with the single source production of NO performed with NEURON simulator[13].Then we investigated the effect of NO in cerebellar learning trough the addition of different background noises. In principle, the incoming CS and US stimuli should exert a depression only at the pf-PC synapses active right before the US stimuli. By adding an increasing noise these learning processes result directly impaired.When including NO-dependent plasticity, we can highlight a different behavior of during a CS and 4 Hz simulation. Here, only the pf-PC synapses receiving the CS stimuli have sufficient NO for plasticity, while the ones randomly activated by noise remain under threshold.
Discussion
The results demonstrate that NO interaction significantly affects synaptic plasticity, dynamically adjusting learning rates based on synaptic activity patterns. This mechanism enhances the cerebellum's capacity to prioritize relevant inputs and mitigate learning interference selectively modulating synaptic efficacy. Our results prove that NO could act as a noise filter, thus focusing learning in the cerebellum only on the relevant inputs for the ongoing task. The NODS implementation connects the molecular processes and large spiking neural network-level learning. This work underscores the critical role of NO in cerebellar function and offers a robust framework for exploring NO-dependent plasticity in computational neuroscience.

Figure 1. Spiking neural network with NODS mechanism. (A) SNN of the cerebellum microcircuit, with the different populations and detail of CS, US and Background Noise stimuli. (B) One trial of the EBCC protocol with timing of the stimuli. (C) The NO production mechanism at a single synapse. (D) NO as volume transmitter at different pf-PC synapses.
Acknowledgements
This research is supported by Horizon Europe Program for Research and Innovation under Grant Agreement No. 101147319 (EBRAINS 2.0). The simulations in NEURON were implemented by Stefano Masoli, Department of Brain and Behavioral Sciences, Università di Pavia, Pavia, Italy.
References
1. https://doi.org/10.1126/science.1470903
2. https://doi.org/10.1016/s0896-6273(00)80340-2
3. https://doi.org/10.1523/JNEUROSCI.4064-13.2014
4. https://doi.org/10.1074/jbc.M111.289777
5. https://doi.org/10.1016/0006-2952(89)90403-6
6. https://doi.org/10.1073/pnas.122206399
7. https://doi.org/10.1016/j.celrep.2016.03.004
8. https://doi.org/10.3389/fnsys.2022.919761
9. https://doi.org/10.3389/fninf.2018.00088
10. https://doi.org/10.1016/j.niox.2009.07.002
11. https://doi.org/10.3389/fninf.2019.00063
12. https://doi.org/10.1109/TBME.2015.2485301
13. https://doi.org/10.1007/978-3-319-65130-9_9

Speakers

Alberto Antonietti

Alessandra Pedrocchi

Carlo Andrea Sartori

Alessandra Maria Trapani

Benedetta Gambosi

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P269: Modeling Calcium-Mediated Spike-Timing Dependent Plasticity in Spiking Neural Networks

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P269 Modeling Calcium-Mediated Spike-Timing Dependent Plasticity in Spiking Neural Networks

Francesco De Santis1*,Carlo Andrea Sartori1*, LeoCottini1, RiccardoMainetti1, MatteoMaresca1, Alessandra Pedrocchi1, Alberto Antonietti1

1 Department of Electronics, Information and Bioengineering,Politecnicodi Milano, Milano, Italy

* Email:francesco.desantis@polimi.it,carloandrea.sartori@polimi.it
Introduction

Calcium dynamics serve asbridge between neuronal activity and synaptic plasticity, orchestrating the biochemical cascades thatdeterminesynaptic strengthening (LTP) or weakening (LTD)[1].Extending thework of Graupner and Brunel [2],Chindemiand colleagues recently introduced a data-constrained model of plasticity based on postsynaptic calcium dynamics in the neocortex[3].Themodel has been developed for NEURON simulationscapturing diverse plasticity dynamics with a single parameter set across pyramidal cell-types. In this work, we translatedChindemi’smodel to a spiking neural network by implementing a point neuron model andaunified synapse, testing it across various calcium-concentration scenarios.

Methods
We developed our model using NESTML[4], an open-source language integrated with NEST[5]simulator, enablingthe application of our models to diverse neural networks. The implemented neuron was built upon the existingHill-Tononi(HT)model, which already incorporates detailed NMDA and AMPA conductance dynamics[6].As inChindemi, the synapse was instead based ontheTsodyks-Markram(TM)stochastic synapse model[7], allowing to manipulatevesicle release probability.Following pairedpre-and post-synaptic activitycalcium-dependent processesinfluencesynaptic efficacyat both sides.Our implementation extends these established components to create a comprehensive framework that captures therelationship between calcium dynamics and synaptic plasticity whilemaintainingcomputational efficiency for network-scale simulations.
Results
We firstvalidatedour model for the TM stochastic synapsepaired withHTmodificationstoaccount forcalcium currentspostsynaptic neuron.Then, we connected two neurons and stimulated either the pre-or post-synapticneuron directly, creatingrespectively NMDA andVDCC calcium currents.Next, we testedthepaired activation of pre-and post-synaptic neurons at varying time intervals.The results of these simulationsare comparable withthe ones ofChindemiet al.Finally, we adjusted LTD and LTP thresholds to match calcium signal properties of pyramidal neurons across different cortical layers. Our simpler point neuron model successfully replicatedfindingsobtained with multicompartmentalmodelswhilemaintainingcomputational efficiency.
Discussion
Our workimplementscalcium-dependent plasticity into an efficientmodel for spiking neurons. Wevalidatedthat our point neuron approach reproduces the complex calcium dynamics and plasticity outcomes across different stimulation patterns. Bymaintainingthe ability to capture layer-specific plasticity with adjusted LTP/LTD thresholds, we preserve biological accuracy while reducing computational demands.Our efficient implementation of calcium-dependent plasticitypossibly enableslarge-scale spiking neural network simulations to study how synaptic mechanisms affect network functionality.

Acknowledgements
The work of AA, AP,CAS, andFDSin this research is supported by Horizon Europe Program for Research and Innovation under Grant Agreement No.101147319 (EBRAINS2.0)andEBRAINS-Italy (European BrainReseArchINfrastructureS-Italy),granted by the Italian National Recovery and Resilience Plan (NRRP), M4C2, funded by the EuropeanUnion –NextGenerationEU(Project IR0000011, CUP B51E22000150006).

References
1.https://doi.org/10.1016/S0959-4388(99)80045-2
2. https://doi.org/10.1073/pnas.1109359109
3.https://doi.org/10.1038/s41467-022-30214-w
4.doi:10.5281/zenodo.12191059
5.https://doi.org/10.4249/scholarpedia.1430
6.https://doi.org/10.1152/jn.00915.2004
7.https://doi.org/10.1073/pnas.94.2.719

Speakers

Alberto Antonietti

Francesco De Santis

Alessandra Pedrocchi

Carlo Andrea Sartori

Leo Cottini

Riccardo Mainetti

Matteo Maresca

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P270: Subthreshold extracellular electric fields alter how neurons respond to naturally occurring synaptic inputs in temporal interference stimulation

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P270 Subthreshold extracellular electric fields alter how neurons respond to naturally occurring synaptic inputs in temporal interference stimulation

Ieva Kerseviciute1, Michele Migliore2, Rosanna Migliore2,Ausra Saudargiene*3, Adam Williamson4

1The Life Sciences Center, Vilnius University, Vilnius, Lithuania
2Institute of Biophysics, National Research Council, Palermo, Italy
3Neuroscience Institute, Lithuanian University of Health Sciences, Kaunas, Lithuania
4St. Anne’s University Hospital, Brno, Czech Republic

*Email: ausra.saudargiene@lsmu.lt

Introduction

Temporal interference (TI) stimulation enables noninvasive and spatially selective neuromodulation of deep brain structures [1,2]. This approach exploits the nonlinear response of neurons to electric fields by delivering multiple kHz-range oscillations, which interfere and generate an effective low-frequency envelope only at the target site [1,2]. This mechanism allows for selective activation of deep neuronal populations without affecting the overlying tissue. Recent studies have successfully applied this stimulation to the human hippocampus, showing significant effects on memory function [3, 4]. Despite its potential for clinical applications, the neural mechanisms underlying TI-induced effects remain poorly understood.

Methods

We used a biophysically accurate computational neuron model to investigate how subthreshold electric fields influence neural activity in the CA1 hippocampal pyramidal neurons. These neurons receive inputs from Schaffer collaterals, known to play an integral role in memory formation. To replicate this connectivity, we implemented AMPA and NMDA synapses at the proximal apical dendrites, with synaptic activity driven by hippocampal CA3 activity recordedin-vivo. The model neuron was placed in a uniform electric field, simulating the effects of an externally applied field between two conducting plates.

Results

Consistent with previously published modelling results [4], we observed that the electric field strength required to elicit action potentials grew with increasing carrier frequency. Moreover, the subthreshold electric field strength also depended on the orientation of the model neuron in the electric field, requiring higher amplitude when the neuron was perpendicular rather than parallel to the direction of the electric field. Following an long-term potentiation (LTP) induction protocol, the subthreshold stimulation affected the synaptic weight distribution by altering the spike timing, firing frequency, and inter-spike interval patterns. A similar effect was observed with naturally occurring synaptic inputs.

Discussion

In summary, our model shows that subthreshold electric fields alter how neurons respond to naturally occurring synaptic inputs by affecting underlying long-term synaptic plasticity processes. The impact of TI on synaptic plasticity may underlie its effects on memory enhancement, observed in human experiments. The stimulation efficacy is partly determined by the neuron orientation in the electric field, as not all neurons are affected equally. Since our study focuses on single-neuron processes, further research is needed to explore network-level effects.

Acknowledgements
●

We acknowledge a contribution from the Italian National Recovery and Resilience Plan (NRRP), M4C2, funded by the European Union – NextGenerationEU (Project IR0000011, CUP B51E22000150006, "EBRAINS-Italy", and support from EU HORIZON-INFRA-2022-SERV-B-01, project 101147319 — EBRAINS 2.0.

References

[1]https://doi.org/10.1016/j.cell.2017.05.024
[2]https://doi.org/10.1126/science.aau4915
[3]https://doi.org/10.1038/s41593-023-01456-8

[4]https://doi.org/10.1101/2024.12.05.24303799

Speakers

Rosanna Migliore

Researcher, Istituto di Biofisica - CNR

Computational NeuroscienceEBRAINS-Italy Research Infrastructure for Neuroscience https://ebrains-italy.eu/

Michele Migliore

Ausra Saudargiene

Adam Williamson

Ieva Kerseviciute

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P274 Dynamics of sensory stimulus representations in recurrent neural networks and in mice

Lars Schutzeichel*1,2,3, Jan Bauer1,4,5, Peter Bouss1,2, Simon Musall3, David Dahmen1and Moritz
Helias1,2

1Institute for Advanced Simulation (IAS-6), Jülich Research Centre, Germany
2Department of Physics, Faculty 1, RWTH Aachen University, Germany
3Institute of Biological Information Processing (IBI-3), Jülich Research Centre, Germany
4Gatsby Unit for Computational Neuroscience, University College London, United Kingdom
5Edmond and Lily Safra Center for Brain Sciences, The Hebrew University of Jerusalem, Israel

*Email: lars.schutzeichel@rwth-aachen.de
Introduction

The information about external stimuli is encoded in the responses of neuronal populations in the brain [1,2], forming neural representations of the stimuli. The diversity of responses is reflected in the extent of these neural representations in neural state space (Fig. 1a). In recent years, understanding the manifold structure underlying neuronal responses [3] has led to insights into representations in both artificial [4] and biological networks [5]. Here, we extend this theory by examining the role of recurrent network dynamics in deforming stimulus representations over time and their influence on stimulus separability (Fig. 1b). Furthermore, we assess the information conveyed for multiple stimuli (Fig. 1c).
Methods
We simulate recurrent networks of binary neurons and study their dynamics analytically using a two-replica mean-field theory, reducing the dynamics of complex networks to only three relevant dynamical quantities: the population rate and the representation overlaps within and between stimulus classes. These networks are fit to Neuropixels recordings from the superior colliculus of awake behaving mice. To assess the information conveyed by multiple stimuli, we analyze the mutual information between an optimally trained readout and the stimulus class. To calculate the overlap of representations within and across stimulus classes, we utilize spin glass methods [6].
Results
Stimulus separability and its temporal dynamics are shaped by the interplay of three dynamical quantities: the mean population activity E and the overlaps θ= and θ≠, which represent response variability within and across stimulus classes, respectively (Fig. 1b). For multiple stimuli, there is a trade-off: as the number of stimuli increases, more information is conveyed, but stimuli become less separable due to their growing overlap in the finite-dimensional neuronal space (Fig. 1c). We find that the experimentally observed small population activity R lies in a regime where information grows extensively with the number of stimuli, sharply separated from a second regime in which information converges to zero.
Discussion
Separability is a minimal requirement for meaningful information processing: The signal propagates to downstream areas, where, along the processing hierarchy, representations of different perceptual objects must become increasingly separable to enable high-level cognition. Our theory reveals that sparse coding not only provides a crucial advantage for information representation but is also a necessary condition for non-vanishing asymptotic information transfer. Our work thus provides a novel understanding of how collective network dynamics shape stimulus separability.

Figure 1. Overview. a: Stimulus representations characterized by their distance from the origin R and their extent θ=. b: Temporal evolution of representations of stimuli from two classes. A linear readout quantifies the separability between the classes of stimuli for every point in time. c: The separability measure also determines the information content in the population signal for P≥2 stimuli.
Acknowledgements
This work has been supported by DFG project 533396241/SPP2205
References
[1]https://doi.org/10.1126/science.3749885
[2]https://doi.org/10.1016/j.tics.2013.06.007
[3]https://doi.org/10.1103/PhysRevX.8.031003
[4]https://doi.org/10.1038/s41467-020-14578-5
[5]https://doi.org/10.1016/j.cell.2020.09.031
[6]https://doi.org/10.1088/1751-8121/aad52e

Speakers

David Dahmen

Lars Schutzeichel

Jan Bauer

Peter Bouss

Simon Musall

Moritz Helias

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P275: A Computational Framework for Investigating the Impact of Neurostimulation on Different Configurations of Neuronal Assemblies

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P275 A Computational Framework for Investigating the Impact of Neurostimulation on Different Configurations of Neuronal Assemblies

Spandan Sengupta*1, 2, Milad Lankarany1, 2, 3, 4, 5

1Krembil Brain Institute, University Health Network, Toronto, ON, Canada
2Institute of Biomedical Engineering, University of Toronto, Toronto, ON, Canada
3Department of Physiology, University of Toronto, Toronto, ON, Canada
4KITE-Toronto Rehabilitation Institute, University Health Network, Toronto, ON, Canada5Center for Advancing Neurotechnological Innovation to Application (CRANIA), Toronto, ON, Canada

*Email: spandan.sengupta@mail.utoronto.ca
Introduction
Pathological oscillations in brain circuits are a known biomarker of neurological disorders, exemplified by increased power in the beta (12-30 Hz) frequency band in Parkinson’s disease [1]. Neurostimulation techniques like Deep Brain Stimulation (DBS) can disrupt pathological oscillations and improve symptoms [2]. However, how and why different stimulation patterns have different impacts on circuits is not fully understood. Recent studies show stimulation-induced biomarkers such as Evoked Resonant Neural Activity (ERNA) [3] associated with stimulation patterns (e.g. frequency) and circuit motifs (e.g. strength of excitatory-inhibitory connectivity) [4]. To study how stimulation patterns impact different circuit motifs, we developed a computational framework to model electrical stimulation on pre and post synaptic activity of neurons embedded in neuronal networks.
Methods
We wanted to study the effect of electrical stimulation, and in particular, the effect of different frequencies during and after stimulation on different circuit motifs. We employed spiking neural networks composed of leaky integrate-and-fire (LIF) neurons combined in a variety of excitatory-inhibitory configurations. To model DBS, we implemented perturbations analogous to electrical stimulation. To further explore how electrical stimulation affects pathological oscillations, we used Brunel’s network [5] tuned to show oscillatory activity at specific frequencies. We aim to study how different patterns of stimulation can suppress these oscillations.

Results
Aligned with experimental findings, our simulations demonstrated that continuous high-frequency electrical stimulation induced more suppression of neuronal activity compared to low-frequency stimulation [6]. In some circuit motifs, we also observed sustained low-frequency oscillatory activity after the high-frequency stimulation had ended (Fig 1B). We aim to characterise the impact of different frequencies on Brunel’s network and their ability to suppress pathological oscillations. We expect to find stimulation patterns that can disrupt these oscillations and qualitatively the circuit activity to a more healthy physiological state. We expect these frequencies to be dependent on the excitatory-inhibitory characteristics of the network.

Discussion
Our study utilizes a simplified model of LIF neurons configured in different motifs, offering a foundational understanding of oscillation modulation with different patterns of electrical stimulation. Future research can expand this model to incorporate more biophysically realistic circuits, such as those found in the hippocampus, critical for memory processing[7], or the basal ganglia, implicated in movement disorders[8]. Investigating these complex circuits will further bridge the gap between computational models and the intricate dynamics of brain networks in health and disease, potentially leading to refined therapeutic strategies.

Figure 1. A: Schematic of a circuit comprised of populations A (exc) and B (inh) that project to O, along with recurrent connections. Electrical stimulation is applied to neurons in A B: Population firing rate during and after 100 Hz electrical stimulation. Dashed red lines indicate that start and end of stimulation. C: Schematic for DBS implementation
Acknowledgements
I would like to thank Dr Frances Skinner (University of Toronto) for her supervision and her help in conceptualising this research idea. I would also to thank Dr Shervin Safavi (Max Planck Institute for Biological Cybernetics) and Dr Thomas Knoesche (Max Planck Institute for Human Cognitive and Brain Sciences) for their help with the modelling and theoretical aspects of this work.

References
●
https://doi.org/10.1152/jn.00697.2006
●
https://doi.org/10.1002/mds.22419
●
https://doi.org/10.1002/ana.25234
●
https://doi.org/10.1016/j.nbd.2023.106019
●
https://doi.org/10.1023/A:1008925309027
●
https://doi.org/10.1016/j.brs.2021.04.022
●
https://doi.org/10.1038/nature15694
●
https://doi.org/10.1016/j.baga.2011.05.001

Speakers

Milad Lankarany

Spandan Sengupta

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P276: Simulation tools for reaction-diffusion modeling

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P276 Simulation tools for reaction-diffusion modeling

Saana Seppälä*1, Laura Keto1, Derek Ndubuaku1, Annika Mäki1, Tuomo Mäki-Marttunen1, Marja-Leena Linne1, Tiina Manninen1

1Faculty of Medicine and Health Technology, Tampere University, Tampere, Finland

*Email: saana.seppala@tuni.fi

Introduction
With advancements in computing power and cellular biology, the simulation of reaction-diffusion systems has gained increasing attention, leading to the development of various simulation algorithms.While we have experience testing separate tools for cell signaling and neuronal networks [1–4], we have not extensively evaluated cell-level tools that integrate both reaction and diffusion algorithms or support co-simulation. This study aims to provide a comprehensive assessment of reaction-diffusion and co-simulation tools, including NEURON [5], ASTRO [6], and NeuroRD [7], to determine their suitability for our research needs.

Methods
Most available reaction-diffusion algorithms and tools are based on partial differential equations or the reaction-diffusion master equation simulated using an extended Gillespie stochastic simulation algorithm [8]. In this study, we implement identical diffusion, reaction, and reaction-diffusion models across selected tools, testing both simple and complex cell morphologies. We conduct simulations, compare results across different tools, and evaluate their consistency. Additionally, we assess the usability and suitability of each tool in various simulation settings, including ease of implementing cell morphologies and equations, computational efficiency, and support for co-simulation.

Results
The simulation algorithms and tools vary significantly in usability and functionality. For instance, some tools support realistic cell morphologies, while others are limited to simplified geometries such as cylinders. Additionally, not all tools allow implementation of reactions involving three reactants, restricting their applicability for certain biological simulations. Despite these differences, a comparison of simulation results across the tools reveals a high degree of similarity, indicating that the underlying models produce consistent outcomes. Furthermore, variations in computational efficiency and ease of implementation are observed, highlighting trade-offs between flexibility, accuracy, and usability across the tools.

Discussion
A thorough understanding of the properties and capabilities of different reaction-diffusion simulation tools is essential for developing more advanced and biologically accurate models. Evaluating these tools provides valuable insights into their strengths and limitations, facilitating the integration of multiple simulation approaches. In particular, this knowledge enables the development of co-simulations that combine reaction diffusion models with spiking network simulations, enhancing the accuracy and scope of computational neuroscience research.

Acknowledgements
This work was supported by the Research Council of Finland (decision numbers 330776, 355256 and 358049), the European Union's Horizon Programme under the Specific Grant Agreement No. 101147319 (EBRAINS 2.0 Project), and the Doctoral School at Tampere University.

References
1.https://doi.org/10.1093/bioinformatics/bti018
2.https://doi.org/10.1155/2011/797250
3.https://doi.org/10.3389/fninf.2018.00020
4.https://doi.org/10.1007/978-3-030-89439-9_4
5.https://doi.org/10.1017/CBO9780511541612
6.https://doi.org/10.1038/s41467-018-05896-w
7.https://doi.org/10.1371/journal.pone.0011725
8.https://doi.org/10.1021/j100540a008

Speakers

Marja-Leena Linne

Tiina Manninen

Tuomo Mäki-Marttunen

Speakers

Masanori Shimono

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P280: Cerebellar neural manifold encoding complex eye movements in 2D

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P280 Cerebellar neural manifold encoding complex eye movements in 2D

Juliana Silva de Deus1, Akshay Markanday2, Erik De Schutter1, Peter Thier2,Sungho Hong*1,3

1Computational Neuroscience Unit, Okinawa Institute of Science and Technology, Okinawa, Japan
2Department of Cognitive Neurology, Hertie Institute, University of Tübingen, Tübingen, Germany
3Center for Cognition and Sociality, Institute for Basic Science, Daejeon, South Korea

*Email: sunghohong@ibs.re.kr

Introduction
Kinematic parameters of our movements, such as velocity and duration, can undergo random and systematic changes, but movement endpoints can be precisely maintained. The cerebellum is well-known for its role in thisfunction[1], but how its neurons concurrently encode several kinematic parameters, necessary for movement precision, has been unknown.We recently identified low-dimensional patterns, called the neural manifold, in the activity of the cerebellar neurons and showed that those multi-dimensional patterns encoded the peak velocity and duration of 1D eye movements, contributing to flexible control of those parameters [2]. In this study, we investigated how those findings can extend to 2D eye movements made in different directions.

Methods
We analyzed the activity of 54 cerebellar Purkinje cells (PC) from the oculomotor vermis in three adult male rhesus monkeys performing two different saccadic eye movement tasks. In the first, the animals made 15° saccades from a fixation point to a visual targetrandomly presented at one of the ten angles(0°-315°, 45° intervals).In the second, they performed a cross-axis adaptation task [3] where initial horizontal jumps of a target from a fixation point were followed by 5° vertical leapsbefore finishing the primary saccades. We analyzed the PC simple spike (SS) activity by identifying its low-dimensional manifold and examining how the manifold varies with the saccade angle and complex spike (CS) firing.

Results

In many PCs (n=39), CSs fired between 100ms and 200 ms after a targetonsetwith well-defined preference for certain target directions (θCS-ON), confirming the directional nature of CS firing for retinalslips[4-6]. We also identified the PC-SS manifold (d=4, explaining >88% variance) for saccadic eye movements with a remarkably simple structure comprising direction-independent latent dynamics and dependent, multi-dimensional gain field, generalizing previous studies [2,7]. How CS and SS firings depend on movement direction (θ) in individual PCs was too heterogeneous to show a clear correlation (P=0.22). However, we found that the gain field looked highly organized for θ-θCS-ONbut much less for θ.

Discussion
Together with our previous study [2], these results show that PC population firing has a remarkably simple structure for representing several kinematic parameters of eye movements, such as velocity, duration, and direction, simultaneously and independently via a low-dimensional neural manifold. Our findings suggest that the cerebellar neural circuit generates neural dynamics optimal for flexible and precise control of complex movements with many degrees of freedom.

Acknowledgements
A.M. and P.T. were supported by DFG Research Unit 1847 “The Physiology of distributed computing underlying higher brain functions in non-human primates.” J.S.D., S.H., and E.D.S. were supported by the Okinawa Institute of Science and Technology Graduate University. S.H. was also supported by the Center for Cognition and Sociality (IBS‐R001‐D2), Institute for Basic Science, South Korea.
References
● https://doi.org/10.1146/annurev-vision-091718-015000
● https://doi.org/10.1038/s41467-023-37981-0
● https://doi.org/10.1007/BF00228022
● https://doi.org/10.1038/33141
● https://doi.org/10.1523/JNEUROSCI.4658-05.2006
● https://doi.org/10.1152/jn.90526.2008
● https://doi.org/10.1038/nature15693

Speakers

Juliana Silva de Deus

Akshay Markanday

Erik De Schutter

Peter Thier

Sungho Hong

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P281: Data-driven biophysically detailed computational modeling of neuronal circuits with the NeuroML standard and software ecosystem

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P281 Data-driven biophysically detailed computational modeling of neuronal circuits with the NeuroML standard and software ecosystem

Ankur Sinha*1, Padraig Gleeson1, Adam Ponzi1, Subhasis Ray2, Sotirios Panagiotou3, Boris Marin4, Robin Angus Silver1

1Department of Neuroscience, Physiology and Pharmacology, University College
2TCG CREST, Kolkata, India
3Erasmus University Rotterdam, Rotterdam, Netherlands
4Universidade Federal do ABC, São Bernardo do Campo, Brazil

*Email: ankur.sinha@ucl.ac.uk

Introduction

Computational models are essential for integrating multiscale experimental data into unified theories and generating new testable hypotheses. Realistic models that include biological intricacies of neurons (morphologies, ionic conductances, subcellular processes) are critical tools for gaining a mechanistic understanding of neuronal processes. Their complexity and the disjointed landscape of software for computational neuroscience, however, makes model construction, fitting to experimental data, simulation, and re-use and dissemination a considerable challenge. Here, we present NeuroML and show that it accelerates modelling workflows and promotes FAIR (Findable, Accessible, Interoperable, Reusable) and Open computational neuroscience[1].

Methods
NeuroML provides two components: a standard and a software ecosystem. The standard is specified by a two part schema. The first part constrains the structure of NeuroML models and is used to validate model descriptions and generate libraries for programming languages. The second part consists of corresponding definitions of the dynamics of model entities in the Low Entropy Modelling Specification language[2] that allows translation of NeuroML models into simulator specific formats. The software ecosystem includes libraries and tools for building and working with NeuroML models in addition to a number of simulation engines and other NeuroML compliant tools that support different stages of the model life cycle.
Results
NeuroML is an established standardised language that provides a simulator independent model representation and accompanying ecosystem of compliant tools that support all stages of the model life cycle: creating, validating, visualising, analysing, simulating, optimising, sharing, re-using models. It provides a curated set of model building blocks for constructing new models and thus also serves as a didactic resource. We demonstrate how NeuroML supports the model life cycle by presenting a number of published NeuroML models in different species (C. elegans, rodents, humans) and different brain regions (cortex, cerebellum), highlighting their scientific contributions. We also list resources on using NeuroML and existing models.
Discussion
NeuroML is a mature standard that has evolved over years of interactions with the computational neuroscience community. The NeuroML community has strong links with simulator development communities to ensure that NeuroML remains up to date with the latest modelling requirements, and that tools remain NeuroML compliant. NeuroML also ensures that it remains extensible to cater to modelling entities that are not yet part of the standard. NeuroML also links to other neuroscience initiatives (PyNN, SONATA[3]), systems biology standards (SBML, SED-ML) and machine learning/AI formats (Model Description Format[4]) to promote interoperability. Finally, a large archive of published standardised models supports re-use of existing models.

Acknowledgements
We thank all members of the NeuroML community who have contributed to the development of the standard and the software ecosystem over the years.
References
● https://doi.org/10.7554/eLife.95135
● https://doi.org/10.3389/fninf.2014.00079
● https://doi.org/10.1371/journal.pcbi.1007696
● https://doi.org/10.1007/s10827-024-00871-5

Speakers

Subhasis Ray

Ankur Sinha

Padraig Gleeson

Adam Ponzi

Sotirios Panagiotou

Bóris Marin

Angus Silver

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P282: Population-level mechanisms of model arbitration in the prefrontal cortex

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P282 Population-level mechanisms of model arbitration in the prefrontal cortex

Jae Hyung Woo1*, Michael C Wang1*, Ramon Bartolo2, Bruno B. Averbeck3,Alireza Soltani1+

1Department of Psychological and Brain Sciences, Dartmouth College, Hanover, NH 03755, USA
2The Laboratory of Sensorimotor Research, National Eye Institute, National Institutes of Health, Bethesda, MD, USA
3Laboratory of Neuropsychology, National Institute of Mental Health, Bethesda, MD 20892, USA
+Email:alireza.soltani@gmail.com
Introduction

One of the biggest challenges of learning in naturalistic settings is that every choice option involves multiple attributes or features, each of which could potentially predict the outcome. To make decisions accurately and efficiently, the brain must attribute the outcome to the relevant features of a choice or action, while disregarding the irrelevant ones. To manage this uncertainty, it is proposed that the brain maintains several internal models of the environment––each predicting outcomes based on different attributes of choice options––and utilizes the reliability of these models to select the appropriate one to guide decision making [1-3].

Methods
To uncover computational and neural mechanisms underlying model arbitration, we reanalyzed data from high-density recordings of the lateral prefrontal cortex (PFC) activity in monkeys performing a probabilistic reversal learning task with uncertainty about the correct model of the environment. We constructed multiple computational models based on reinforcement learning (RL) to fit choice behavior on a trial-by-trial basis, which allowed us to infer animals’ learning and arbitration strategies. We then used estimates based on the best-fitting model to identify single-cell and population-level neural signals related to learning and arbitration in the lateral PFC.

Results
We found evidence of dynamic, competitive interactions between stimulus-based and action-based learning, alongside single-cell and population-level representations of the arbitration weight. Arbitration enhanced task-relevant variables, suppressed irrelevant ones, and modulated the geometry of PFC representations by aligning differential value axes with the choice axis when relevant and making them orthogonal when irrelevant. Reward feedback emerged as a potential mechanism for these changes, as reward enhanced the representation of relevant differential values and choice while adjusting the alignment between differential value and choice subspaces according to the adopted learning strategy.

Discussion
Overall, our results shed light on two major mechanisms for the dynamic interaction between model arbitration and value representation in the lateral PFC. Moreover, they provide evidence for a set of unified computational and neural mechanisms for behavioral flexibility in naturalistic environments, where there is no cue that explicitly signals the correct model of the environment.

Acknowledgements
None
References

1.https://doi.org/10.1038/s41386-021-01123-1
2.https://doi.org/10.1016/j.neubiorev.2020.10.022
3.https://doi.org/10.1038/s41386-021-01108-0

Speakers

Bruno Averbeck

Alireza Soltani

Jae Hyung Woo

Michael Wang

Ramon Bartolo

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P283: A biophysical model of CA1 pyramidal cell heterogeneity in memory stability and flexible decision-making

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P283 A biophysical model of CA1 pyramidal cell heterogeneity in memory stability and flexible decision-making

Fei Song*1,2, Bailu Si3,4

1State Key Laboratory of Robotics, Shenyang Institute of Automation Chinese Academy of Science, Shenyang, Liaoning, China
2University of Chinese Academy of Sciences, Beijing, Beijing, China
3School of Systems Sciences, Beijing Normal University, Beijing, Beijing, China
4Chinese Institute for Brain Research, Beijing, Beijing, China

*Email: songfeo20160903@gmail.com
Introduction

The entorhinal-hippocampal system is essential for spatial memory and navigation. CA1 integrates spatial and non-spatial inputs via two pathways: the perforant path (PP) and temporoammonic path (TA), processed by pyramidal cells (PCs) [1]. We propose a biophysical model with simple PCs and complex PCs [2]. Simulations in novel environments (Fig. 1a) show sPCs maintain stable spatial coding, while cPCs integrate spatial and attentional inputs, supporting decision-making. In familiar settings (Fig. 1b), cPCs adapt to changes while sPCs preserve stable encoding, enabling memory retention and comparison of past and new experiences. This model unifies CA1’s roles in memory and decision-making.

Methods
We model CA1 as a two-layer network: deep-layer sPCs receive MEC input, while superficial-layer cPCs integrate MEC and LEC signals. Synaptic plasticity follows Hebbian learning, with SC weights adapting via dendritic-somatic co-activation and TA weights via rate-dependent learning, constrained by proximal-distal gradients. Simulations include a 10m track and 5m open-field, where MEC provides grid-cell input, LEC encodes egocentric cues, and CA3 supplies place-cell activity [3,4]. Memory recovery is evaluated via place field stability (JS distance), while stimulus-specific information quantifies spatial and attentional encoding variability [5]. A population decoder (MLP) predicts location and attention from CA1 activity.
Results
CA1 supports flexible decision-making by integrating spatial and perceptual information. In novel environments, sPCs ensure spatial stability, while cPCs encode stimulus-specific cues. A proximal-distal gradient in cPCs appears with fixed cues but disappears with moving cues, confirming their adaptive role. Population decoding shows cPCs excel in attention tracking, while sPCs maintain spatial coding. CA1 also aids memory updating. When CA3 recall is incomplete, CA1 preserves past memories longer than expected, slowing decay. When TA introduces novelty, cPCs encode new inputs while sPCs retain old ones, enabling stable yet adaptive memory processing. This mirrors real-world experiences, such as recognizing familiar but altered locations.
Discussion
Our model captures CA1 neuron heterogeneity and projection preferences in decision-making and memory updating. However, it simplifies CA3’s proximodistal heterogeneity, where pattern separation (proximal) and completion (distal) may influence CA1 dynamics [6]. Future work should refine CA3 input representation. CA1’s dual-pathway structure aligns with cognitive map theory, where novel environments require integration, while familiar ones involve consolidation. This parallels the Tolman-Eichenbaum Machine (TEM) model of hippocampal function [7]. The dual-pathway structure may reflect a generalized neuronal computation mechanism, extending beyond navigation and memory to broader cognitive functions.

Figure 1. Fig. 1 Functional Framework of the Hippocampus. (a) CA1 supports flexible decision-making in novel environments by integrating sensory inputs and generating context-specific representations. (b) CA1 facilitates memory updating in familiar environments by comparing stored memories with current experiences.
Acknowledgements
Not applicable.
References
● https://doi.org/10.1038/nn.2894
● https://doi.org/10.1038/nn.4517
● https://doi.org/10.1016/j.neucom.2020.10.013
● https://doi.org/10.1007/BF00237147
● https://api.semanticscholar.org/CorpusID:10081513
● https://doi.org/10.1371/journal.pbio.2006100
● https://doi.org/10.1016/j.cell.2020.10.024

Speakers

Fei Song

Bailu Si

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P284: Subthalamic LFP Spectral Decay Captures Movement-Related Differences Between Parkinson’s Disease Phenotypes

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P284 Subthalamic LFP Spectral Decay Captures Movement-Related Differences Between Parkinson’s Disease Phenotypes

Luiz Ricardo Trajano da Silva1, Maria Sheila Guimarães Rocha2, Slawomir Nasuto3, Bradley Voytek4, Fabio Godinho5,Diogo Coutinho Soriano*1

1Center of Engineering, Modeling and Applied Social Sciences, Federal University of ABC (UFABC), São Bernardo do Campo, Brazil
2Department of Neurology, Santa Marcelina Hospital, São Paulo, Brazil
3University of Reading, Berkshire, United Kingdom
4Department of Cognitive Science, Halıcıoğlu Data Science Institute, University of California, San Diego, La Jolla, CA, USA
5Division of Neurosurgery, Department of Neurology, Hospital das Clínicas, University of São Paulo Medical School, São Paulo, Brazil

*Email: diogo.soriano@ufabc.edu.br
Introduction

Parkinson’s Disease (PD) is a heterogeneous neurodegenerative disorder characterized by a wide range of motor and non-motor symptoms [1]. Movement disorder specialists classify PD into subtypes, including tremor dominant (TD) and postural instability and gait disorder (PIGD) [2, 3, 4]. One promising robust biomarker for deep brain stimulation (DBS) therapy is the 1/f^chi spectral decay observed in local field potentials (LFPs). This decay has been linked to the excitatory/inhibitory synaptic balance, providing valuable insights into neuronal circuit dynamics [5, 6, 7, 8]. Therefore, this study explores changes in the spectral decay across rest and movement conditions in different PD phenotypes, aiming to advance personalized DBS strategies.
Methods
STN-LFP recordings from 35 hemispheres (15 TD, 20 PIGD) during rest and movement (elbow extension and flexion) conditions (1 minute each) were acquired during the intraoperative procedure for implanting DBS electrodes. Welch periodogram and spectral parametrization, as proposed in [5], were used to estimate the LFP adjusted low beta (13–22 Hz), high beta (22–35 Hz) rhythms bandpower (i.e., corrected by 1/f^chi background), and the spectral decay parameter chi. Mixed-ANOVA was used to evaluate differences between subtypes and rest/movement conditions. The procedure was approved by the ethical committee for research in human beings(CAAE: 62418316.9.2004.0066).
Results
(Fig. 1) shows the parametrized spectral decay for TD (A) and PIGD (B) phenotypes, the respective PSDs adjusted by 1/f^chi (panels C and D), and the box plots for the bandpower rhythms and spectral decay (E, F, G). Lower beta power showed an interaction between phenotype and motor condition (F(1,33) = 6.67, p = 0.014), with a significant decrease during movement (p = 0.003) for the TD group. High beta bandpower showed a marginal effect for phenotype during rest (F(1,33) = 3.39, p = 0.07). The spectral decay exponent also indicates an interaction between phenotype and the motor condition (F(1,33) = 5.67, p = 0.02), with a post-hoc analysis unveiling a marginal phenotype difference during movement (p = 0.088).
Discussion
Spectral parameterization revealed significant differences between the TD and PIGD subtypes, highlighting distinct neuronal dynamics in the subthalamic nucleus (STN) during movement (elbow flexion). Our findings indicate that beta-band suppression during movement, as documented in previous studies [9–12], is predominantly driven by TD patients. Conversely, the PIGD group showed increased high-beta activity, which has been linked to motor rigidity symptoms [13], along with a steeper aperiodic exponential decay, suggesting a more inhibited synaptic balance in the STN during movement. These results highlight the potential of spectral decay components as biomarkers for personalized DBS strategies for PD patients.

Figure 1. Figure 1 – Aperiodic-adjusted and Aperiodic Component PSDs and Grouped Boxplot for subtype and rest/movement conditions. A and B, aperiodic component PSD for TD and PIGD groups, respectively. C and D, Aperiodic-adjusted PSDs for TD and PIGD groups, respectively. E, F, and G, Boxplot for subtype and rest/movement conditions exhibiting mixed-ANOVA results. (.) 0.05 < 𝘱 < 0.1 ;*𝘱 < 0.05;**𝘱 < 0.01
Acknowledgements
Authors acknowledge the financial support of the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - Brasil (CAPES)- Finance Code 001 and CNPq (grant number 313970/2023-8).
References
1.https://doi.org/10.1038/s41582-021-00486-9
2.https://doi.org/10.1016/S0140-6736(21)00218-X
3.https://doi.org/10.1002/acn3.312
4.https://doi.org/10.1016/j.parkreldis.2019.05.024
5.https://doi.org/10.1038/s41593-020-00744-x
6.https://doi.org/10.1038/s41531-018-0068-y
7.https://doi.org/10.1016/j.neuroimage.2017.06.078
8.https://doi.org/10.1523/JNEUROSCI.2041-09.2009
9.https://doi.org/10.1016/j.expneurol.2012.05.013
10.https://doi.org/10.1093/brain/awh106
11.https://doi.org/10.1002/mds.10358
12.https://doi.org/10.1093/brain/awf135
13.https://doi.org/doi: 10.1002/mds.26759

Speakers

Diogo Soriano

Luiz Ricardo Trajano da Silva

Maria Rocha

Slawomir Nasuto

Bradley Voytek

Fabio Godinho

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P285: A two-region recurrent neural network reproduces the cortical dynamics underlying subjective visual perception

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P285 A two-region recurrent neural network reproduces the cortical dynamics underlying subjective visual perception

Artemio Soto-Breceda1, Nathan Faivre2, João Barbosa3,4,Michael Pereira1

1.Univ. Grenoble Alpes, Inserm, Grenoble Institut Neurosciences, 38000 Grenoble, France
2.Univ. Grenoble Alpes, Univ. Savoie Mont Blanc, CNRS, LPNC, 38000 Grenoble, France
3 Cognitive Neuroimaging Unit, INSERM, CEA, CNRS, Université Paris-Saclay, NeuroSpin center, Gif/Yvette, France
4 Institut de Neuromodulation, GHU Paris Psychiatrie et Neurosciences, Centre Hospitalier Sainte-Anne, Université Paris Cité, Paris, France
Introduction

This study aims to model the cortical activity associated with the detection of visual stimuli, as well as the subjective duration of visual percepts and associated confidence. We propose a two-region neural network model: a sensory region integrating sensory inputs and a decision region with longer integration timescales. The model is constrained by biological parameters to simulate region-dependent temporal integration and includes top-down feedback and excitation-inhibition balance to test hypotheses on the neural basis of perception.

Methods
The model consists of a recurrent rate-based neural network of excitatory (80%) and inhibitory (20%) neurons with GABA, AMPA, and NMDA synapses. The sensory region receives and integrates sensory inputs and projects to a decision region with longer integration timescales. This decision region defines whether and when a near-threshold stimulus is detected. The dynamics of the simulated neural activity in the sensory region were compared to the dynamics of neural activity consisting of local field potentials recorded using stereotaxic EEG in humans undergoing epilepsy monitoring and associated behavioral measures of detection, response times, subjective confidence and subjective duration collected with a time-reproduction task [1].
Results
The model successfully replicated key behavioral metrics. Qualitatively, simulated activity in the decision region matched high-gamma activity recorded in the anterior insula, while sensory region activity aligned with activity in the inferior temporal cortex during a face detection task. We find that, for example, temporal integration in sensory regions explains the magnitude-duration illusion, where higher intensity stimuli are perceived as longer. We also examined model predictions when altering the E/I ratio by changing the synaptic strength of NMDA receptors in either the excitatory or inhibitory population [2], or modulating the top-down feedback. We intend to test alternative models corresponding to different hypotheses on how temporal integration explains subjective aspects of perception such as duration and confidence.
Discussion
Many studies have provided computational models of perceptual decision-making. However, the neuronal mechanisms underlying the subjective aspects of perception remain poorly understood. Here, starting from a model of decision-making [3], we harness temporal properties of these subjective aspects of perception to isolate the underlying neuronal mechanism. The model is able to predict behavior in perceptual decision-making tasks. This model allows us to investigate how biological parameters such as E/I balance or top-down feedback affect behavior and cortical activity during perceptual decision-making tasks. We will interpret our findings in the context of current theories of consciousness.

Acknowledgements
-
References
● https://doi.org/10.1101/2024.03.20.585198
● https://doi.org/10.1523/JNEUROSCI.1371-20.2021
● https://doi.org/10.1016/S0896-6273(02)01092-9

Speakers

Artemio Soto Breceda

Nathan Faivre

João Barbosa

Michael Pereira

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P286: Learning in Visual Cortex: Sparseness, Balance, Decorrelation, and the Parameters of the Leaky Integrate-and-Fire Model.

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P286 Learning in Visual Cortex: Sparseness, Balance, Decorrelation, and the Parameters of the Leaky Integrate-and-Fire Model.

Martin J. Spencer1*, Marko A. Ruslim1*, Hinze Hogendoorn2, Hamish Meffin1, Yanbo Lian1, Anthony N. Burkitt1,3

1Department of Biomedical Engineering, University of Melbourne, Melbourne, Victoria 3010, Australia
2School of Psychology and Counselling, Queensland University of Technology, Kelvin Grove, Queensland 4059, Australia
3Graeme Clark Institute for Biomedical Engineering, University of Melbourne, Melbourne, Victoria 3010, Australia

*Equal first authors. Email:martin.spencer@unimelb.edu.au
InIntroduction:Sparseness is a known property of information representation in the cortex [1]. A sparse neural code represents the underlying causes of sensory stimuli and is resource efficient [2]. Computational models of sparse coding in the visual cortex typically use an objective function with an information maximization term and a neural activity minimization term, a top-down approach [3]. In contrast, this study trained a spiking neural network using Spike-Timing-Dependent Plasticity (STDP) learning rules [4]. The resulting sparseness, decorrelation, and balance in the network was then quantified; a bottom-up approach [5]. To confirm the mechanisms of sparseness, results were replicated across 3 models of increasing complexity.
Methods:A biologically grounded V1 model was made up of separate populations of excitatory and inhibitory Leaky Integrate and Fire (LIF) neurons with all-to-all connectivity via delta-current synapses. Input was provided by Poisson neurons with spike rates representing the output of separate ON and OFF neurons calculated using a centre-surround whitening filter applied to natural images.
The V1 LIF neuron spike rates were maintained at a target rate using a homeostatic threshold adjustment. Synaptic weights were adjusted using a triplet STDP rule [4] for the excitatory-excitatory neuron synapses and a symmetric STDP rule for other connections. Learning was normalised using subtractive normalisation and multiplicative normalisation.
Results:Training was performed using 1200 batches of 100 natural image patches for 400 ms each (~11 hours). There were 512 LGN neurons (256 ON, 256 OFF) and 500 V1 neurons (400 Excitatory, and 100 Inhibitory). The network was found to achieve a sparse representation. The level of sparseness was found to depend on the parameters of the LIF model. These mechanisms were additionally explored in a simple single neuron model and computationally efficient smaller model (Figure 1). Decorrelation was observed to result from the weights chosen by STDP. ‘Loose’ and ‘tight’ balance was confirmed using comparison of the relative strength of excitatory and inhibitory input.
Discussion:In the biologically grounded V1 model the results showed that the balance was maintained across long (~1 s) and short (~10 ms) times scales. Where pairs of neurons had receptive fields with high correlations it was found that there was correlated high mutual inhibition leading to diversity and information maximization in the network.
In all 3 models, higher sparseness (ς) was caused by lower output spike rates in the LIF neurons (Figure 1 A and C, efficient model). In the efficient and biologically grounded models this was associated more Gabor-like receptive fields (Figure 1 B and D). Other parameters of the LIF model were also examined, including membrane time constant, input spike rate and number of inputs.

Figure 1. Figure 1: (A) Sparseness (ς) measured in the computationally efficient V1 neuron model of 64 neurons with a 5 Hz target mean spike rate. (B) Associated normalised synaptic weight weights to 9 V1 neurons from the ON (red) and OFF (blue) input neurons. (C-D) 30 Hz target mean spike rate.
Acknowledgements
Acknowledgements

This work was supported by an Australian Research Council Discovery Grant (DP220101166).
References
References
[1] - https://doi.org/10.1038/s41467-020-14645-x
[2] - https://doi.org/10.1038/srep17531
[3] - https://doi.org/10.1038/381607a0
[4] -https://doi.org/10.1523/JNEUROSCI.1425-06.2006
[5] - https://doi.org/10.1101/2024.12.05.627100

Speakers

Hamish Meffin

Anthony Burkitt

Martin Spencer

Marko Ruslim

Yanbo Lian

Hinze Hogendoorn

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P287: Pulsatile and direct current stimulation in functional network model of decision making

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P287 Pulsatile and direct current stimulation in functional network model of decision making

Cynthia Steinhardt*1,2,Paul Adkisson3, Gene Fridman3
1 Simons Society of Fellows, Junior Fellow, New York, New York 10010
2 Center for Theoretical Neuroscience, Zuckerman Brain Science Institute, Columbia University, New York, New York 10027
3 Department of Biomedical Engineering, Johns Hopkins School of Medicine, Baltimore, Maryland 21287
*Email: cs4248@columbia.edu

Introduction
Pulsatile stimulation has been the primary method for neural implants in sensory restoration and neuropathology treatment (e.g., Parkinson’s, epilepsy) since the first neural implant [1]. Recently, non-invasive transcranial direct/alternating current (DC/AC) stimulation has gained interest, offering broader accessibility without surgery. However, to be viable, effective non-invasive alternatives must match or exceed the efficacy of implants in modulating neural circuits. Pulsatile and DC stimulation effects in complex networks have not been directly compared due to the need for detailed biophysical models. We address this gap.
Methods
Our prior work showed that pulsatile stimulation alters firing patterns in single neurons in complex ways depending on pulse parameters and spontaneous activity [3]. Similarly, we modeled and characterized the effects of DC stimulation on single neurons [4]. Here, we extend these models, modifying linear-integrate-and-fire (LIF) models to include approximations of these effects so that we can accurately simulate local stimulation in a 1000-neuron network. We simulate pulsatile and DC stimulation at equivalent local dosing levels and at behaviorally equivalent levels and compare network effects in a winner-take-all decision-making circuit for motion detection.
Results
The network processes moving dots and determines whether the majority are moving left or right. We identified pulse rates for suprathreshold pulses that match DC stimulation’s effects on the firing rate in the left-motion detection part of the network. At this level, pulsatile stimulation induced a stronger, faster bias toward leftward decisions. When matched for behavioral bias, pulsatile stimulation resisted feedback inhibition and had conflicting effects with recurrent feedback. DC stimulation, in contrast, propagated through the network more strongly due to recurrent excitation but was more affected by feedback inhibition [5].
Discussion
This study provides the first direct comparison of how pulsatile and DC stimulation influence network activity up to the behavioral-level, using accurate approximations of electrical stimulation. We show that these two forms of stimulation interact differently with network dynamics, suggesting different therapeutic applications. Additionally, we present open-access tools for modeling, which could enhance patient-specific disease models. These tools allow for mechanistic insights beyond the LIF and threshold models currently used.

Acknowledgements
Acknowledgments
We thank the Simons Society of Fellows (965377), Gatsby Charitable Trust (GAT3708), Kavli Foundation, and NIH (R01NS110893) for support.
References
References
1.Loeb, G. E. (2018). Neural Prosthetics.Appl Bionics Biomech,2018, 1435030.
2.Giordano, J., et al. (2017). Mechanisms of tDCS.Dose-Response,15(1), 1559325816685467.
3.Steinhardt, C. R., et al. (2024). Pulsatile stimulation disrupts firing.Nat Commun,15(1), 5861.
4.Steinhardt, C. R., & Fridman, G. Y. (2021). DC effects on afferents.iScience,24(3).
5.Adkisson, P. W., et al. (2024). Galvanic vs. pulsatile effects.J Neural Eng,21(2), 026021.

Speakers

Cynthia Steinhardt

Paul Adkisson

Gene Fridman

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P288: Modelling the temperature profile of the retina in response to nanophotonic neuromodulation

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P288 Modelling the temperature profile of the retina in response to nanophotonic neuromodulation

Daniel B Richardson1, James Begeng1, Paul R Stoddart1,Tatiana Kameneva*1,2

1Department of Engineering Technologies, School of Engineering, Swinburne University of Technology, Australia

2Iverson Health Innovation Institute, Swinburne University of Technology, Australia

*Email: tkam@swin.edu.au

Electrical stimulation of neurons has been used as a reliable technique to elicit action potentials in implantable devices. Recently, novel optical stimulation techniques have been developed as alternatives to electrical stimulation. One approach involves applying near infrared wavelengths of light to stimulate neurons. Neurophotonic stimulation may increase the resultant visual acuity compared to electrical stimulation as it does not apply any current and thus has no current spread. As a result of applying nanophotonic stimulation, the retina experiences an increase in temperature. For this reason, modelling the temperature profile within the retina is vital in testing the feasibility of optical stimulation techniques.
Step 1: To model the temperature profile in a retina environment, a Monte Carlo simulation was implemented in MATLAB. The environment consisted of four layers: water, gold nanorods, retinal tissue, and a layer of glass. A 750nm beam was used to simulate near infrared stimulation at varying powers that matched the experimental values of Begeng et al (2023). Each layer had specified coefficients obtained from literature, which included the absorption and scattering coefficients, scattering anisotropy, volumetric heat capacity, and thermal conductivity. The simulation models the temperature profile through finite element modelling of the defined geometry. It determines the temperature through tracking the photon paths of the stimulation beam, monitoring how it progresses through the tissues via their varying scattering coefficients and refractive indexes. It then models the florescence and absorption of the tissues through probabilistic determination. Theamountof photons absorbed, and its associated power, is then used in conjunction with the heat equation to determine the temperature.
Step 2: A single-compartment Hodgkin-Huxley model of a temperature-sensitive rat RGC was constructed in the NEURON simulation environment. The model uses the Gouy-Chapman-Stern theory of temperature-variant bilayer capacitance, and experimentally-derived temperature dependence for key sodium, potassium, calcium and leak ion channels, as well as cytosolic resistance. Thermal profiles for the pulse durations were approximated using the thermal model described in Step 1.

The simulation temperature model demonstrated general agreement with the experimental results, showing comparable peak temperatures and maintaining a consistent trend with the varying pulse durations. Furthermore, the proposed temperature model allows for estimation of the temperature profile on the retinal surface, which is difficult to measure experimentally. Hodgkin-Huxley model replicated the main features of nanophotonic stimulation, including an initial driving subthreshold depolarisation hump, followed by an action potential, inhibition and excitation phenomena, that were dependent on the pulse duration.

Acknowledgements
-
References
Begeng JM, Tong W, Rosal B, Ibbotson M, Kameneva T, Stoddart PR (2023) Activity of retinal neurons can be modulated by tunable near-infrared nanoparticle sensors, ACS Nano 17 (3), 2079 – 2088,doi/10.1021/acsnano.2c07663

Speakers

Paul Stoddart

Gabriel Ocker

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P295: How baseline activity determines neural entrainment by transcranial alternating current stimulation (tACS) in recurrent inhibitory-excitatory networks

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P295 How baseline activity determines neural entrainment by transcranial alternating current stimulation (tACS) in recurrent inhibitory-excitatory networks

Saeed Taghavi*1,2, Gianluca Susi1, Alireza Valizadeh1,2,Fernando Maestú1

1Zapata-Briceño Institute of Neuroscience, Madrid, Spain
2Physics Department, Institute for Advanced Studies in Basic Sciences (IASBS), Zanjan, Iran

*Email: saeed.taghavi.v@gmail.com

Introduction
Neuronal oscillations play a key role in cognition and can be modulated by transcranial alternating current stimulation (tACS). However, the mechanisms underlying network-level entrainment remain unclear. We investigate how a balanced excitatory-inhibitory network of adaptive exponential integrate-and-fire neurons responds to sinusoidal stimulation. We analyze phase-locking to determine how external rhythmic inputs influence neural synchronization at different baseline network states.
Methods
We simulate a recurrent EI network that receives Poisson-distributed background input. Three baseline synchronization levels are studied, reflecting the degree of natural synchronization in neuronal activity within the network before any external stimulation is applied. Additionally, tACS-like stimulation is applied at frequencies ranging from 5 to 60 Hz with five different amplitudes (3, 4, 6, 8, and 10). Each condition is repeated over nine trials to ensure reliability. To quantify network entrainment, we compute the phase-locking value between the population activity and the stimulation. Furthermore, we calculate the spike-field coherency of individual neurons and measure changes in SFC with and without stimulation to assess how neuronal firing aligns with the external signal.
Results
Our results show that baseline network synchrony strongly influences entrainment. Networks with higher intrinsic synchrony exhibit stronger phase locking with the stimulation. When the stimulation frequency is close to the endogenous frequency, PLV increases with stimulation amplitude, suggesting that stronger inputs enhance entrainment only when the stimulation frequency matches the endogenous frequency. Frequency-dependent effects emerge, with the most robust responses occurring near the network’s intrinsic oscillation frequency. Individual neurons display varying phase coherence, with some aligning strongly to the stimulation while others remain weakly affected.
Discussion

We discovered that tACS-induced neural entrainment behaves in a way that challenges conventional expectations. While you might assume higher baseline synchrony leads to broader entrainment, we found the opposite. Networks with low baseline synchrony actually exhibit broader locking across a wider range of external frequencies. Conversely, highly synchronized networks show stronger locking, but it's tightly confined to the baseline frequency's vicinity. This counterintuitive result underscores the delicate balance between baseline synchrony and tACS effectiveness, highlighting the need for nuanced approaches in cognitive and therapeutic applications.

Figure 1. (a) Entrainment of population activity to tACS varies with network synchrony and stimulation strength. Higher synchrony or amplitude increases peak PLV but narrows the high-PLV region. (b) Stimulation does not significantly alter firing rates but enhances phase coherence. (c) Spike phase coherence changes shows a peak when stimulation matches network frequency.
Acknowledgements

References
[1]https://doi.org/10.1016/j.heliyon.2024.e41034
[2]https://doi.org/10.1101/2023.05.19.541493
[3]https://doi.org/10.1016/j.neuroimage.2022.118953
[4]https://doi.org/10.3390/biomedicines10102333
[5]https://doi.org/10.3389/fnsys.2022.827353

Speakers

Saeed Taghavi

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P296: The modelling of the action potentials in myelinated nerve fibres

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P296 The modelling of the action potentials in myelinated nerve fibres
K. Tamm 1*, T. Peets 1, J. Engelbrecht 1,2

1Tallinn University of Technology, Department of Cybernetics, Tallinn, Estonia
2Estonian Academy of Sciences, Tallinn, Estonia
*kert.tamm@taltech.ee

Introduction. The classical Hodgkin-Huxley (HH) model [1] describes the propagation of an axon potential (AP) in unmyelinated axons. In many cases, the axons have a myelin sheath. A theoretical model is proposed describing the AP propagation in myelinated axons drawing inspiration from the studies of Lieberstein [2], who included the possible effect of inductance. The Lieberstein-inspired model (in the form of coupled partial differential equations (PDEs)) can describe all the essential effects characteristic to the formation and propagation of an AP in an unmyelinated axon. Then a phenomenological model for a myelinated axon is described including the influence of the structural properties of the myelin sheath and the radius of an axon.

Methods. The model equations are solved numerically by making use of the pseudospectral method (PSM) [3]. Briefly, the main point of PSM is that the discrete Fourier transform (DFT) can be used to approximate space derivatives reducing, therefore, the PDE to an ordinary differential equation (ODE) and then to use standard ODE solvers for integration in time. Here the ODE solver is used through its NumPy (a Python package) implementation. The parameters for the model are collected from experiments (most of them from the classical HH paper [1]) or estimated separately based on experimental observations.

Results. Using the parameters from the experiments we investigated the numerical solutions of the noted model for the unmyelinated axon and demonstrated that the behaviour of the solutions is in the physiologically plausible range and the key characteristics of the nervous signalling are fulfilled (annihilation of counter-propagating signals, threshold, refractory period). The model includes the structural properties of the myelin sheath: the μ-ratio (longitudinal geometry) and g-ratio (perpendicular geometry). The key difference between the classical HH model and the Lieberstein-inspired model used here is that the mechanism for signal propagation along the axon emerges like a wave as a consequence of opting to keep the inductivity.

Discussion. The goal of constructing yet another equation for describing the AP propagation along the axon is clearer physical interpretation as we start from the elementary form of Maxwell equations which is modified to include the influence of myelination on the signal propagating along the axon. It is important to stress that the proposed continuum-based model is philosophically similar to how the transmission line equations are composed. The ‘unit-cell’ in the context of the myelinated axon in the model is composed of the node of Ranvier and the myelinated section next to it. Having a pair of PDEs with a straightforward connection to underlying physics could be useful for investigating causal connections in the context of nerve signalling.

Acknowledgements
This research was supported by the Estonian Research Council (PRG 1227). Jüri Engelbrecht acknowledges the support from the Estonian Academy of Sciences.
References
[1] https://doi.org/10.1113/jphysiol.1952.sp004764
[2] https://doi.org/10.1016/0025-5564(67)90026-0
[3] https://doi.org/10.1007/978-3-030-75039-8

Speakers

Tanel Peets

Kert Tamm

Jüri Engelbrecht

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P297: Autonomous Generation of Neuronal Connection by Axon Guidance Factors

P300: The neuron-synapse dynamics control correlational timescales in the developing visual system

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P300 The neuron-synapse dynamics control correlational timescales in the developing visual system

Ruben A. Tikidji-Hamburyan1*, Matthew T. Colonnese1
1School of Medicine. and Health. Sciences, the George Washington Univ., Washington, D.C., USA

*Email: rath@gwu.edu

Introduction: During early development, the retinal spontaneous wave-like activity provides positional (spatial) information, encoded in coarse-grained (>100 ms) inter-neuron spike correlations [1], needed for the refinement of retinothalamic, but also for thalamocortical and intracortical connections. The formation of sub- and cortical networks goes in parallel with the refinement of the retinothalamic connections, therefore, spatial information must be transferred by an unrefined, imprecise thalamic network. Thalamocortical relay neurons (TCs) receive 10 to 20 inputs from neighboring ganglion cells at this age [2,3] , which should cause fast (<100ms) timescale correlations in TCs firing. Here, we model how these correlational timescales are regulated.
Methods: TC neurons were simulated as two compartments (dendrosomatic:NaF, KDr, NaP, CaL, CaT, KA, SK, H currents and Ca2+ dynamics and axonal:NaF,KDr) conductance-based model derived from an adult [4]. The parameters were fitted to reproduce the dynamics of mouse TCs recorded at postnatal day 7 (P7) by genetic algorithms with nondominated sorting [5] and with Krayzman’s adaptive multiobjective optimization [6]. The network model consists of 120 TC neurons activated by spikes of retinal ganglion cells (rGCs) recorded ex-vivo at P6-P9. The probability of connections and the synaptic weights are modeled as Gaussian dependence on the distance. Each synapse was modeled by 2-stages: presynaptic depression and postsynaptic NMDAR and AMPAR currents [2,7].
Results: We show that with synaptic convergences observed at P7, either adult neuronal dynamics or synaptic current composition causes fast timescale correlations and a dramatic decrease in spatial information encoded in TC spikes. Therefore, we call them “parasitic” correlations. However, parasitic correlations are suppressed independently of convergence if the model replicates P7 neuronal dynamics and dominance of slow NMDAR currents - the landmark property at this age [3]. Moreover, the interplay between neuron and synaptic dynamics suppresses only parasitic correlation, keeping informative slow timescale correlations intact. In contrast, parasitic correlations are negligible in networks with adult convergence and don’t need to be suppressed.
Discussion: Our results suggest that developing neurons regulate their membrane and synaptic dynamics to preserve information critical for proper circuit formation by suppressing non-informative parasitic correlations. As we showed, parasitic correlations can be invariantly suppressed, while informative correlation passes through an unrefined and imprecise network. Our modeling opens critical general questions: how are correlations transferred, and how does a network regulate correlational timescales? The answers to these questions go beyond just neuronal excitability, as for synchrony transferring [8], and require synergetic regulation of both neuronal and synaptic dynamics.

Acknowledgements
This work was supported by R01EY022730 and R01NS106244
References
● 10.1523/JNEUROSCI.19-09-03580.1999
● 10.1002/cne.22223
● 10.1016/S0896-6273(00)00166-5
● 10.1371/journal.pcbi.1006753
● 10.1109/4235.996017
● 10.7554/eLife.84333
● 10.1523/JNEUROSCI.4276-07.2008
● 10.1016/j.neuron.2013.05.030

P307: A Computational Model to Study Effects of Hidden Hearing Loss in Noisy Environments

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P307 A Computational Model to Study Effects of Hidden Hearing Loss in Noisy Environments

Siddhant Tripathy*1, Maral Budak2, Ross Maddox3, Gabriel Corfas3, Michael T. Roberts3, Anahita H. Mehta3, Victoria Booth4, Michal Zochowski1,5

1Department of Physics, University of Michigan, Ann Arbor, USA
2Department of Microbiology and Immunology, University of Michigan, Ann Arbor, USA
3Kresge Hearing Research Institute, University of Michigan, Ann Arbor, USA
4Department of Mathematics, University of Michigan, Ann Arbor, USA
5Biophysics Program, University of Michigan, Ann Arbor, USA

*Email: tripps@umich.edu

Introduction

Hidden Hearing Loss (HHL) is an auditory neuropathy leading to reduced speech intelligibility in noisy environments despite normal audiometric thresholds. One of the leading hypotheses for such degraded performance is myelinopathy, a permanent disruption in the myelination patterns of type 1 Spiral Ganglion Neuron (SGN) fibers [1,2]. Previous studies on location discriminability in the Medial Superior Olive (MSO) cells in the left and right hemispheres as a function of the interaural time difference (ITD), have shown that myelinopathy leads to signatures of HHL [3]. However, the effects of noise on location discriminability is unknown.
Methods
To investigate these effects, we developed a physiologically based model that incorporates SGN fiber activity to sound stimuli processed through a peripheral auditory system model [4]. To simulate myelinopathy, we introduced random variations in the position of myelination heminodes, which generates phase shifts in the spike timing of affected fibers. To test the subsequent effects on sound localization, we constructed a network model that simulates the propagation of SGN responses to cochlear nuclei and the MSO populations. We varied the location of the sound impulse by introducing a phase shift in the input in one ear relative to the other, with background noise signals kept stationary.
Results
Upon adding noise to the sound stimuli, we find that spikes in a given SGN fiber's spike train are shifted inhomogeneously, leading to a reduction in phase locking of single fibers to sound. The effects of myelinopathy on population behavior are thus more pronounced in the presence of noise. Subsequently in the localization network, we find that the sensitivity to ITD is reduced in myelinopathy conditions, and that this effect is significantly exacerbated when we introduce noisy background stimuli, a signature of HHL.
Discussion
We find that noisy environments exacerbate HHL symptoms. This model may be useful in understanding the downstream impacts of SGN neuropathies.

Acknowledgements
This research was supported in part by National Institute of Health grant: NIH MH135565 (MZ and ST) and R01DC000188 (GC).R01DC000188
References
● https://doi.org/10.1038/ncomms14487
● Budak, M., Grosh, K., Sasmal, A., Corfas, G., Zochowski, M., and Booth, V. (2021). Contrasting mechanisms for hidden hearing loss: Synaptopathy vs myelin defects. PLoS Comput. Biol. 17:e1008499. doi: 10.1371/journal.pcbi.1008499
● Budak, M., Roberts, M. T., Grosh, K., Corfas, G., Booth, V. and Zochowski, M. (2022). Binaural Processing Deficits Due to Synaptopathy and Myelin Defects. Front. Neural Circuits 16:856926. doi: 10.3389/fncir.2022.856926
● https://doi.org/10.1121/1.1453451PMID: 12051437.

Speakers

Siddhant Tripathy

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P308: Brain-Inspired Recurrent Neural Network Featuring Dendrites for Efficient and Accurate Learning in Classification Tasks

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P308 Brain-Inspired Recurrent Neural Network Featuring Dendrites for Efficient and Accurate Learning in Classification Tasks

Eirini Troullinou*1,2, Spyridon Chavlis1, Panayiota Poirazi1

1Institute of Molecular Biology and Biotechnology, Foundation for Research, and Technology-Hellas, Heraklion, Greece
2Institute of Computer Science, Foundation for Research, and Technology-Hellas, Heraklion, Greece

*Email: eirini_troullinou@imbb.forth.gr

Introduction

Artificial neural networks (ANNs) have achieved substantial advancements in addressing complex tasks across diverse domains, including image recognition and natural language processing. These networks rely on a large number of parameters to attain high performance; however, as the complexity of ANNs increases, the challenge of training them efficiently also escalates [1]. In contrast, the biological brain, which has served as a fundamental inspiration for ANN architectures [2], exhibits remarkable computational efficiency by processing vast amounts of information with minimal energy consumption [3]. Moreover, biological neural networks demonstrate robust generalization capabilities, often achieving effective learning with limited training samples, a phenomenon known as few-shot learning.

Methods
In an effort to develop a more biologically plausible computational model, we propose a sparse, brain-inspired recurrent neural network (RNN) that incorporates biologically motivated connectivity principles. This approach is driven by the computational advantages of dendritic processing [4], which have been extensively studied in biological neural networks. Specifically, our model enforces structured connectivity constraints that emulate the physical relationships between dendrites, neuronal somata, and inter-neuronal connections. These biologically inspired connectivity rules are implemented via structured binary masking, thereby regulating the network's architecture based on empirical neurophysiological observations.

Results
To assess the efficacy of the proposed model, we conducted a series of experiments on benchmark image and time-series datasets. The results indicate that our brain-inspired RNN attains the highest accuracy achieved by a conventional (vanilla) RNN while utilizing fewer trainable parameters. Furthermore, when the number of trainable parameters is increased, our model surpasses the peak performance of the vanilla RNN by a margin of 3–20%, depending on the dataset. In contrast, the conventional RNN exhibits overfitting tendencies, leading to significant performance degradation.

Discussion
In summary, we present a biologically inspired RNN architecture that incorporates dendritic processing and sparse connectivity constraints. Our findings demonstrate that the proposed model outperforms traditional RNNs in both image and time-series classification tasks. Additionally, the model achieves competitive performance with fewer parameters, highlighting the potential role of dendritic computations in machine learning. These results align with experimental evidence suggesting the critical contribution of dendrites to efficient neural processing, thereby offering a promising direction for future ANN development.

AcknowledgementsThis work was supported by the NIH (GA: 1R01MH124867-04), the TITAN ERA Chair project under Contract 101086741 within the Horizon Europe Framework Program of the European Commission, and the Stavros Niarchos Foundation and the Hellenic Foundation for Research and Innovation under the 5th Call of Science and Society "Action Always strive for excellence – Theodoros Papazoglou" (DENDROLEAP 28056).
References
[1] Abdolrasol, M. G, et al. (2021). Artificial neural networks based optimization techniques: A review. Electronics, 10(21), 2689.
[2] Sejnowski, T. J. (2020). The unreasonable effectiveness of deep learning in artificial intelligence. PNAS, 117(48), 30033-38.
[3] Attwell, D., & Laughlin, S. B. (2001). An energy budget for signaling in the grey matter of the brain. J Cereb Blood Flow Metab, 21(10), 1133-1145.
[4] Poirazi, P., & Papoutsi, A. (2020). Illuminating dendritic function with computational models. Nat Rev Neurosci, 21(6), 303-21.

Speakers

Spyridon Chavlis

Panayiota Poirazi

Eirini Troullinou

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P312 Modeling Effects of Norepinephrine on Respiratory Neurons

Sreshta Venkatakrishnan*1, Andrew Kieran Tryba2, Alfredo J. Garcia, 3rd3, and Yangyang Wang1

1Department of Mathematics, Brandeis University, Waltham, MA, USA
2Department of Pediatrics, Section of Neurology, The University of Chicago, Chicago, IL, USA
3Institute for Integrative Physiology, The University of Chicago, Chicago, IL, USA

*Email: sreshtav@brandeis.edu

Introduction
The preBötzinger complex (pBC) within the mammalian brainstem,comprised ofintrinsically bursting and spiking neurons,generates the neural rhythm that drives the inspiratory phase of respiration. Norepinephrine (NE), a neuromodulator, differentially modulates synaptically isolated pBC neurons [1]. In cadmium (Cd)-insensitive N-bursting neurons, NE stimulates burst frequency without affecting burst duration. In Cd-sensitive C-bursting neurons, NE increases duration while minimally affecting frequency. NE also induces conditional bursts in tonic spiking neurons, while silent neurons remain inactivein the presenceof NE. In this work, we propose a novel mechanism to simulate the effects of NE in single pBC neurons.

Methods
The pBC neuron model we consider is a single-compartment dynamical system with Hodgkin-Huxley-style conductances, incorporating membrane potential and calcium dynamics, adapted from previous works [2,3,4]. Of particular interest to us amongst the ionic currents incorporated in this model are two candidate burst-generating currents: Cd-insensitive persistent sodium current (INaP) and Cd-sensitive calcium-activated nonspecific cationic current (ICAN). Building on previous efforts for modeling NE via modulating ICAN[2,3] and from experimental evidence in [5], we propose that NE application in the model also leads to an increase in the flux of [Ca2+] between the cytosol and the ER, modeled via inositol-triphosphate, IP3.
Results
The most important finding of this study is the identification of potential mechanisms underlying the NE-mediated induction of tonic spiking neurons to CAN-dependent bursting. Our model predicts that this conditional bursting requires an increase in both IP3and ICAN. This mechanism also induces an increase in N-burst frequency and C-burst duration, while N-burst duration remains unaltered. While modulatingICANincreases C-burst frequency in our model, the opposing effect brought bymodulating IP3effectivelycounters this increase and maintains frequency. Furthermore, we also identify discrete parameter regimes where silent neurons continue to remain inactive in NE. These results are consistent with [1].
Discussion
Conditional bursting has been previously described in rhythmic networks; however, the underlying mechanisms are often unknown. Our model predicts a new mechanism involving NE-signaling, elevating both IP3and ICANin a subset of pBC neurons. These predictions need to be experimentally tested by blocking either IP3or ICAN, and testing whether subsequent NE modulation can no longer recruit this subset of pBC neurons to burst. Moreover, while our model predictions for bursting neurons mostly agree with the experiments in [1], we also notice some discrepancies with respect to burst frequency and duration. Further investigation is required to analyze and understand these disparities.

Acknowledgements
This work has been supported byNIH R01DA057767:CRCNS: Evidence-based modeling of neuromodulatory action on network properties,granted to Yangyang Wang (PI)andAlfredo Garcia at UChicago.
References
[1]https://doi.org/10.1152/jn.01308.2005
[2]https://doi.org/10.1007/s10827-010-0274-z
[3]https://doi.org/10.1007/s10827-012-0425-5
[4]https://doi.org/10.1063/1.5138993
[5]https://doi.org/10.1152/ajpendo.1985.248.6.E633

Speakers

Sreshta Venkatakrishnan

Andrew Tryba

Alfredo Garcia

Yangyang Wang

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P313: Individual differences in neural representations of face dimensions: Insights from Super-Recognisers

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P313 Individual differences in neural representations of face dimensions: Insights from Super-Recognisers

Martina Ventura*1, Tijl Grootswagers1,3, Manuel Varlet1,2, David White2, James D. Dunn2, Genevieve L. Quek1

1The MARCS Institute for Brain, Behaviour and Development, Western Sydney University, Sydney, Australia
2School of Psychology, Western Sydney University, Sydney, Australia
3School of Computer, Data and Mathematical Sciences, Western Sydney University, Sydney Australia
4School of Psychology, The University of New South Wales, Sydney, Australia

*E-mail:martina.ventura@westernsydney.edu.au

Introduction
Face processing is crucial for social interaction, with faces conveying information about identity, emotions, sex, age, and intentions [1]. Recent research has revealed significant individual differences in face recognition ability, with some people displaying exceptional face recognition skills – defined as Super-Recognisers [2,3].However, the brain mechanisms underpinning their superior ability remain unknown, including whether their exceptional face recognition is restricted to identity or also extends to other face dimensions such as sex and age.

Methods
Here we use Electroencephalography (EEG) to investigate the neural processes underlying face dimensions representations in Super-Recognisers (N = 12) and Typical-Recognisers (N = 17).We recorded 64 channel EEG while participants saw 400 naturalistic face images (40 distinct identities stratified by sex, age, and ethnicity) in a rapid 5 Hz randomized stream. We used Multi-Variate Pattern Analysis to measure the strength and temporal dynamics of neural encoding of different facial dimensions in both Super-and Typical-Recognisers.

Results
Our results showed that face identity decoding was stronger forSuper-RecognisersthanTypical-Recognisersstarting around 300ms - a time window typically associated with identity-related processing. In contrast, no differences were found between groups’ decoding profiles for face age, face sex, or face ethnicity.
Discussion
These results suggest that theSuper-Recognisersadvantage may be limited to face identity processing, rather than reflecting a general advantage in face dimension processing. These findings provide a crucial first step toward understanding the neural mechanisms underlying their exceptional face recognition ability.

Acknowledgements
We sincerely appreciate the time and effort of all the participants in this study. Your willingness to take part was essential in making this research possible. Thank you for your valuable contribution.
References
1.Tsao, D. Y., & Livingstone, M. S. (2008). Mechanisms of face perception.Annual review of neuroscience,31, 411–437.https://doi.org/10.1146/annurev.neuro.30.051606.094238

2.Russell, R., Duchaine, B., & Nakayama, K. (2009). Super-recognizers: people with extraordinary face recognition ability.Psychonomic bulletin & review,16(2), 252–257.https://doi.org/10.3758/PBR.16.2.252

3.Dunn, J. D., Summersby, S., Towler, A., Davis, J. P., & White, D. (2020). UNSW Face Test: A screening tool for super-recognizers.PloS one,15(11), e0241747. https://doi.org/10.1371/journal.pone.0241747

Speakers

Martina Ventura

Tijl Grootswagers

Manuel Varlet

David White

James Dunn

Genevieve Quek

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P314: Neural compensation drives functional resilience in a cerebellar model of schizophrenia

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P314 Neural compensation drives functional resilience in a cerebellar model of schizophrenia

Alberto A. Vergani*1, Pawan Faris1, Claudia Casellato1, Marialaura De Grazia1and Egidio U. D'Angelo1,2

1Department of Brain and Behavioral Sciences, University of Pavia, Pavia, Italy
2Digital Neuroscience Center, IRCCS Mondino Foundation, Pavia, Italy

*Email: albertoarturo.vergani@unipv.it

Introduction

Schizophrenia (SZ) affects ~1% of the global population (~24 million) [1]. While cortical and subcortical alterations are well-documented, the cerebellum’s role in cognitive dysfunction (CAS) remains underexplored [2]. SZ-related cerebellar degeneration involves neuron loss, reduced dendritic complexity, and weakened connectivity [3], often countered by compensatory hyperconnectivity [4-8]. Following the 'cognitive dysmetria' hypothesis [9], this study quantifies structural and functional changes in a cerebellar network model under atrophy and compensatory synaptogenesis [10-11].

Methods
Using Brain Scaffold Builder (BSB, [12]), we implemented an atrophy algorithm in a mouse cerebellum model, modulating cellular and network changes via the Atrophy Factor (AF, 0–60%).By preserving electrical cell properties, it simulated schizophrenic neurodegeneration while ensuring anatomical plausibility. Atrophy induced morphological shrinkage, dendritic pruning, radius reduction, neural density loss, and cortical thinning. Changes were quantified via apoptosis, dendritic complexity index (DCI, [13]), and connectivity metrics. Compensation via synaptogenesis increased synapse count with AF. The altered connectome (~30K neurons, EGLIF, [14]) was simulated in NEST [15] under baseline conditions (4 Hz mossy fiber stimulation) to assess firing rate changes.

Results
Atrophy altered network structure, reducing neurons, dendritic complexity, connectivity, and synapse count. Compensation offset this by increasing synapses in survived neuron pairs. Functional changes emerged from structural alterations, with excitability rising, reversing at ~10% AF, and zero-crossing at ~25% AF. Granule and Golgi cells showed opposite trends, while Purkinje, stellate, and basket cells were similar in firing change. DCN-I neurons gradually reduced activity, with compensation lightly delaying decline. DCN-P exhibited the highest resilience until ~25% AF, where compensation collapsed, triggering a firing surge disrupting output to telencephalon.

Discussion
This study examined cerebellar network degeneration while preserving electrical properties, highlighting structural changes, synaptic reorganization, and atrophy-related firing dynamics.Synaptic compensation mitigates pathology-driven neuronal damage, with a transition from hyper- to hypo-excitability, particularly in DCN-P, resembling Stern’s inflection point in neurodegenerative resilience [16]. Future work will explore atrophy-compensation effects on stimulus decoding and learning (eye blink conditioning, [17]), integrate with The Virtual Brain [18], compare with MEA recordings [19], and test therapeutic strategies like TMS and pharmacological interventions to enhance cognitive reserve.

Acknowledgements
Work supported by #NEXTGENERATIONEU (NGEU) and funded by MUR, National Recovery and Resilience Plan (NRRP), project MNESYS (PE0000006) – A Multiscale integrated approach to the study of the nervous system in health and disease (DN. 1553 11.10.2022). The VBT Project has received funding from the European Union's Research and Innovation Program Horizon Europe under grant agreement No 101137289.
References
1 10.1001/jamapsychiatry.2019.3360
2 10.3389/fncel.2024.1386583
3 10.1016/j.biopsych.2008.01.003
4 10.1093/schbul/sbac120
5 10.1038/s41398-023-02512-4
6 10.1016/j.pscychresns.2018.03.010
7 10.1016/j.schres.2022.12.041
8 10.1038/s41386-018-0059-z
9 10.1093/oxfordjournals.schbul.a033321
10 10.1007/s12311-019-01091-9
11 10.1523/JNEUROSCI.0379-23.2023
12 10.1038/s42003-022-04213-y
13 10.1038/s42003-023-05689-y
14 10.3389/fninf.2018.00088
15 10.5281/ZENODO.4018718
16 10.1016/j.neurobiolaging.2022.10.015
17 10.3389/fpsyt.2015.00146
18 10.1093/nsr/nwae079

19 10.1371/journal.pcbi.1004584

Speakers

Claudia Casellato

Alberto Arturo Vergani

Research Fellow, University of Pavia

Egidio D'Angelo

Marialaura De Grazia

Pawan Faris

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P315: Towards brain scale simulations using NEST GPU

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P315 Towards brain scale simulations using NEST GPU

José Villamar*1,2, Gianmarco Tiddia3, Luca Sergi3,4, Pooja Babu1,5, Luca Pontisso6, Francesco Simula6, Alessandro Lonardo6, Elena Pastorelli6, Pier Stanislao Paolucci6, Bruno Golosio3,4, Johanna Senk1,7

1Institute for Advanced Simulation (IAS-6), Jülich Research Centre, Jülich, Germany
2RWTH Aachen University, Aachen, Germany
3Istituto Nazionale di Fisica Nucleare, Sezione di Cagliari, Monserrato, Italy
4Dipartimento di Fisica, Università di Cagliari, Monserrato, Italy
5Simulation and Data Laboratory Neuroscience, Jülich Supercomputing Centre, Jülich Research Centre, Jülich, Germany
6Istituto Nazionale di Fisica Nucleare, Sezione di Roma, Roma, Italy
7Sussex AI, School of Engineering and Informatics, University of Sussex, Brighton, United Kingdom
*Email:j.villamar@fz-juelich.de

Introduction

Efficient simulation of large-scale spiking neuronal networks is important for neuroscientific research, and both the simulation speed and the time it takes to instantiate the network in computer memory are key factors. NEST GPU is a GPU-based simulator under the NEST Initiative written in CUDA-C++ that demonstrates high simulation speeds with models of various network sizes on single-GPU and multi-GPU systems [1,2,3]. On the path toward models of the whole brain, neuroscientists show an increasing interest in studying networks that are larger by several orders of magnitude. Here, we show the performance of our simulation technology with a scalable network model across multiple network sizes approaching human cortex magnitudes.
Methods
For this, we propose a novel method to efficiently instantiate large networks on multiple GPUs in parallel. Our approach relies on the deterministic initial state of pseudo-random number generators (PRNGs). While requiring synchronization of network construction directives between MPI processes and a small memory overhead, this approach enables dynamical neuron creation and connection at runtime. The method is evaluated through a two-population recurrently connected network model designed for benchmarking an arbitrary number of GPUs while maintaining first-order network statistics across scales.
Results
The benchmarking model was tested during an exclusive reservation of the LEONARDO Booster cluster. While keeping constant the number of neurons and incoming synapses to each neuron per GPU, we performed several simulation runs exploiting in parallel from 400 to 12,000 (full system) GPUs. Each GPU device contained approximately 281 thousand neurons and 3.1 billion synapses. Our results show network construction times of less than a second using the full system and stable dynamics across scales. At full system scale, the network model was composed of approximately 3.37 billion neurons and 37.96 trillion synapses (~25% human cortex).

Discussion
To conclude, our novel approach enabled network model instantiation of magnitudes nearing human cortex scale while keeping fast construction times, on average of 0.5s across trials. The stability of dynamics and performance across scales obtained in our model is a proof of feasibility paving the way for biologically more plausible and detailed brain scale models.

Acknowledgements
ISCRA for awarding access to the LEONARDO supercomputer (EuroHPC Joint Undertaking) via theBRAINSTAIN - INFN Scientific Committee 5 project, hosted by CINECA (Italy); HiRSE_PS, Helmholtz Platform for Research Software Engineering - Preparatory Study (2022-01-01 - 2023-12-31), the Horizon Europe Grant 101147319, Joint lab SMHB; FAIR CUPI53C22001400006 Italian PNRR grant.
References
●
https://doi.org/10.3389/fncom.2021.627620
●
https://doi.org/10.3389/fninf.2022.883333
●
https://doi.org/10.3390/app13179598

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P316: Characterization of thalamic stimulus on a cortical column

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P316 Characterization of thalamic stimulus on a cortical column

Pablo Vizcaíno-García*1,2,3, Fernando Maestú1,3, Alireza Valizadeh1 Gianluca Susi1,2,3

1Zapata-Briceño Institute for Human Intelligence, Madrid, Spain.
2Department of Structure of Matter, Thermal Physics and Electronics, School of Physics, Complutense University of Madrid, Madrid, Spain
3Center for Cognitive and Computational Neuroscience, Complutense University of Madrid, Madrid, Spain

*Email: pabvizca@ucm.es

Introduction

Cortical columns are fundamental organizational units in cerebral cortical processing and development [1]. They regularly receive external stimuli, coming from both higher-order areas and the thalamus. Different hypotheses have been proposed regarding the function of the thalamus: it is considered to act as a generator of the alpha rhythm [2]; and it is also said to play a potential role in sensory gating. In this work we focus on exploring the latter process. We investigate how stimuli propagate from one layer of the cortical column to the entire unit, examining how alpha and gamma rhythms may be disrupted or enhanced across the different layers. We build upon the design of a cortical column by Potjans & Diesmann [3].

Methods
We implemented an interconnected set of full-spiking cortical columns, each column encompassing 80000 neurons and 0.3 billion synapses. The connections have been derived from experimental data, utilising diffusion magnetic resonance imaging data. The column’s background stimulus was modified in order to start in a high-coherence state, which more easily allows the characterisation of the response. Said characterisation has been done by injecting a pulse packet into L4E, and obtaining order PRC (Phase Response Curves) which characterise the delays produced by the same stimulus if injected into different phases of the activity [4]. A 1ms wide stimulus was injected into different phases of the gamma period of L4E.
Results
The phases were identified after a gamma band filter (45-80Hz) and applying the Hilbert transform to the cortical column in the absence of a stimulus. The resulting PRC curves can be observed in Fig. 1. This figure presents both the raster plot of a stimulated cortical column and the resulting PRC curve. Each dot of the figure represents one spike of one neuron, in the appropriate time, and the superimposed line is the gamma-filtered population activity. From this figure we can observe there is a sudden halt in the gamma band after stimulation in L4E. The PRC curve was computed as an ensemble average of 10 trials. From this figure we highlight L23E as the only population which is consistently delayed by the input stimulus.
Discussion
From these early results, two main facts have become evident. First, the burst suppression phenomena emerges as a response to stimulating L4E, a layer that mostly receives its inputs from the thalamic nuclei. Second, the PRC of L23E shows the biggest time lag. Another avenue to explore is the variation of these curves in a less coherent network state. This work will seek to elucidate the mechanisms behind both of these phenomena, applying and comparing the results with the well-studied thalamocortical feedback loop. The investigation will contribute to a better understanding of the cortical column dynamics, but additionally will help clarify the effects of the communication between the thalamus and the cortical cortex.

Figure 1. Left: Raster plot of cortical column activity . Each dot represents a spike, and each colour a neuronal population. Imposed over the plot is the activity of each population, computed using a gaussian window over spike times, and normalised for the plot. Right: Phase response curve. Measures the time where each population reaches the first maximum in activity after stimulation injection into L4E.
Acknowledgements
This work was supported by Zapata-Briceño Intstitute of Science.
References
1.https://doi.org/10.1016/B978-0-12-814411-4.00005-6
2. hhtps://doi.org/10.34734/FZJ-2023-02822
3.https://doi.org/10.1093/cercor/bhs358
4.https://doi.org/10.3389/fninf.2010.00006

Speakers

Pablo Vizcaíno

Alireza Valizadeh

Fernando Maestú

Gianluca Susi

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P317: Critical dynamics improve performance in deep learning

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P317 Critical dynamics improve performance in deep learning

Simon Vock*1,2,3,4,5, Christian Meisel1,2,4,5,6

1Computational Neurology Lab, Department of Neurology, Charité – Universitätsmedizin, Berlin, Germany
2Berlin Institute of Health, Berlin, Germany
3Faculty of Life Sciences, Humboldt University Berlin, Germany
4Bernstein Center for Computational Neuroscience, Berlin, Germany
5NeuroCure Cluster of Excellence, Charité – Universitätsmedizin, Berlin, Germany
6Center for Stroke Research, Berlin, Germany

*Email: simon.vock@charite.de
Introduction

Deep neural networks (DNNs) have revolutionized AI, yet their vast parameter space makes training difficult, often leading to inefficiencies or failure. Their optimization remains largely heuristic, relying on trial-and-error design [1,2]. In biological networks, recent evidence suggests that critical phase transitions - balancing signal propagation to avoid die-out or runaway excitation - are key for effective learning [3,4]. Inspired by this, we analyze 80 modern DNNs and uncover a fundamental link between performance and criticality, unifying diverse architectures under a single theoretical perspective. Building on this, we propose a novel training approach that guides DNNs toward criticality, enhancing performance on multiple datasets.
Methods
We characterize criticality in DNNs using three key metrics: A maximum dynamic range Δ [5], a branching parameter σ=1 [6], and the largest Lyapunov exponent λ₀=0 [7]. Our statistical analysis employs multiple tests including Spearman's rank, Wilcoxon signed-rank, Mann-Whitney U, and linear mixed-effects models. We investigate 80 highly optimized DNNs from TorchVision pre-trained on the ImageNet-1k dataset [8]. We use the Modified National Institute of Standards and Technology (MNIST) dataset, a standard benchmark for computer vision. Building on our findings, we develop a novel training objective that specifically drives models toward criticality during the training process.
Results
We derive a set of measures quantifying the distance to criticality on DNNs and analyze 80 pre-trained DNNs from Torchvision (ImageNet-1k). We found that over the last decade, as test accuracies increased, networks became significantly more critical. Our analysis shows that test accuracies are highly correlated with criticality and model size. A linear mixed-effects model shows that distance to criticality and model size explain 60% of the variance in accuracy (R²). A novel training objective that penalizes distance to criticality improves MNIST accuracy by up to 0.8% compared to highly optimized DNNs. In a continual learning setting using ImageNet, this approach enhances neuronal plasticity and outperforms established training techniques.
Discussion
Analyzing 80 diverse DNNs developed over the last decade, we uncover two key ingredients for high-performance deep learning: Network size and critical neuron dynamics. We find that modern deep learning techniques implicitly enhance criticality, driving recent advancements in the field. We show how improved DNN architectures and training approaches promote criticality, and further introduce a novel training method that enforces criticality during training. This significantly boosts accuracy on MNIST. Additionally, our method enhances the network’s plasticity, improving adaptability to new information in continual learning. We expect these findings to generalize to other models and tasks, offering a path toward more efficient AI.

Acknowledgements

References
1. Glorot, X., & Bengio, Y. (n.d.). Understanding the difficulty of training deep feedforward neural networks. Proceedings of the 13th International Conference on Artificial Intelligence and Statistics (AISTATS), 9.
2.https://doi.org/10.1038/nature14539
3.https://doi.org/10.1523/JNEUROSCI.23-35-11167.2003
4.https://doi.org/10.1016/0167-2789(90)90064-V
5.https://doi.org/10.1523/JNEUROSCI.3864-09.2009
6.https://doi.org/10.1103/PhysRevLett.94.058101
7.https://doi.org/10.1103/PhysRevLett.132.057301
8.https://doi.org/10.1109/CVPR.2009.5206848

Speakers

Simon Vock

Christian Meisel

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P318: Algorithmic solutions for spike-timing dependent plasticity in large-scale network simulations with long axonal delays

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P318 Algorithmic solutions for spike-timing dependent plasticity in large-scale network simulations with long axonal delays

Jan N. Vogelsang*1,2, Abigail Morrison*2,3, Susanne Kunkel1

1 Neuromorphic Software Ecosystems (PGI-15), Jülich Research Centre, Jülich, Germany
2RWTH Aachen University, Aachen, Germany
3Institute for Advanced Simulation (IAS-6), Jülich Research Centre, Jülich, Germany

*Email: j.vogelsang@fz-juelich.de
Introduction

The precise timing of neuronal communication is a cornerstone in understanding learning and synaptic plasticity. Spike-timing dependent plasticity (STPD) models in particular rely on the precise temporal difference between pre- and post-synaptic spikes to adjust synaptic strength, where both the diverse axonal propagation delays and dendritic backpropagation delays play a crucial role in determining the precise timing. However, neural simulators, such as NEST, have traditionally represented transmission delays between neurons as a single aggregate delay value because of algorithmic challenges. We present two simulation frameworks addressing these challenges and validate across a set of small- to large-scale benchmarks.

Methods
The NEST simulator reference implementation currently treats the entire delay as dendritic, which allows performing synaptic strength adjustments immediately after the occurrence of a pre-synaptic spike, avoiding costly buffering of spikes. This is an acceptable approximation for small networks but leads to inaccuracies when modeling long-range connections. In this framework, introducing axonal delays causes causality issues. At the time a pre-synaptic spike occurs, post-synaptic spikes only occurring in future time steps might reach the synapse before such spike due to predominant axonal delays. In order to mitigate this issue, one must either correct the weight on later occurrence of such post-synaptic spikes or postpone the STDP update.
Results
Both approaches were implemented and rigorously benchmarked in terms of runtime efficiency and memory footprint for varying synaptic delays and delay partitions. Correcting faulty synaptic updates achieves exceptional performance for fractions of axonal delay equal to or lower than the corresponding dendritic one. Only in the case of predominant and long axonal delays it starts to be outperformed by the alternative approach, which however required fundamental changes to the simulation framework to enable efficient buffering of individual spikes at the synapse level. However, benchmarks show that the approach induces a negative impact on performance for simulations not involving STDP dynamics, unlike the correction-based approach.
Discussion
Although different axonal and dendritic contributions are known to bias the synaptic drift towards either systematic potentiation or depression, there is a lack of simulation studies investigating the effects on network dynamics and learning in large neuronal systems. The ability to differentiate between axonal and dendritic delays represents a significant advance in neural simulation technology, as it addresses a long-standing limitation in spike-timing dependent plasticity modeling in large-scale, distributed simulations and enables future research in learning and plasticity, in particular, investigations of brain-scale models with STDP faithfully representing heterogeneous long axonal delays between areas.

Acknowledgements
I want to thank Dennis Terhorst and Anno Kurth for assistance in benchmarking and running all the required jobs on the HPC systems.
References
-

Speakers

Jan Vogelsang

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P319: Interactions between functional microcircuits involving three inhibitory interneuron subtypes for the surround modulation in V1

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P319 Interactions between functional microcircuits involving three inhibitory interneuron subtypes for the surround modulation in V1

Nobuhiko Wagatsuma*1, Tomoki Kurikawa2, Sou Nobukawa3,4

1Faculty of Science, Toho University, Funabashi, Chiba, Japan
2Future University Hakodate, Hakodate, Hokkaido, Japan
3Department of Computer Science, Narashino, Chiba Institute of Technology, Chiba, Japan
4Department of Preventive Intervention for Psychiatric Disorders, National Institute of Mental Health, National Center of Neurology and Psychiatry, Kodaira, Tokyo, Japan

*Email: nwagatsuma@is.sci.toho-u.ac.jp

Introduction
A functional microcircuit of V1 for interpreting the external world resides in layers 2/3 and consists of excitatory pyramidal (Pyr) neurons and three inhibitory interneuron subtypes: parvalbumin (PV), somatostatin (SOM), and vasoactive intestinal polypeptide (VIP). Recent physiological and computational studies suggest a structured organization of this microcircuit and distinct roles of inhibitory interneuron subtypes in modulating neural activity for visual perception [1,2]. Interactions between these microcircuits across receptive fields are crucial for integrating larger visual regions and forming perception, yet the precise structures and interneuron subtypes mediating these interactions remain unclear.

Methods
We developed a computational microcircuit model of the functional unit with biologically plausible visual cortical layers 2/3 that combined excitatory Pyr neurons and three inhibitory interneuron subtypes and explored the role of specific inhibitory interneuron subtype for mediating the interactions between these two microcircuits via the lateral inhibition across receptive fields (Fig.(A)). We assumed that the receptive fields of these units, which share common orientation selectivity, are spatially adjacent in the visual field. In this study, two functional microcircuits interacted each other via the lateral inhibition from excitatory Pyr neurons in one unit to PV or SOM inhibitory interneurons in the other.
Results
We performed simulations of the model with inputs mimicking small and large visual stimuli used in the physiological experiment [3]. We assumed that the small stimulus was confined to the receptive field of a single unit, whereas the large stimulus extended across the receptive fields of two microcircuits. Model simulations with the large visual stimulus implied that the lateral inhibition from Pyr neurons in one microcircuit to SOM interneurons in the other preferentially induced neuronal firing at beta (13-30 Hz) frequency, in agreement with physiological responses for the surround suppression in V1 [3]. By contrast, the model with the lateral inhibition mediated by PV interneurons distinct modulation patterns from physiological results.
Discussion
Our model reproduced characteristic neuronal activities in V1 induced by the surround modulation when the lateral inhibition across the receptive fields was mediated by SOM interneurons. Our results of model simulations suggested the specific role of SOM interneurons in the long-range lateral interactions across receptive fields in V1, which might contribute to the generation of surround modulation.

Figure 1. (A) Proposed microcircuit model. These two microcircuits interacted each other via lateral connections from Pyr neurons in one unit to PV or SOM interneurons in the other. (B) Simulation results. Black and blue lines indicated the oscillatory responses of the model with the lateral inhibition mediated by SOM and PV interneurons, respectively. The red line was those with the small stimulus.
Acknowledgements
This work was partly supported by the Japanese Society for the Promotion of Science (JSPS) (KAKENHI grants 22K12138, 22K12183, 23H03697, and 23K06394) and a grant of the Research Initiative Program of Toho University (TUGRIP).
References
1. https://doi.org/10.1093/cercor/bhac355
2. https://doi.org/10.1016/j.celrep.2018.10.029.
3. https://doi.org/10.1038/nn.4562.

Speakers

Tomoki Kurikawa

Nobuhiko Wagatsuma

Sou Nobukawa

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P320: Updating of spatial memories in a systems-model of vector trace cells in the subiculum

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P320 Updating of spatial memories in a systems-model of vector trace cells in the subiculum

Fei Wang*1, Andrej Bicanski1

1Department ofPsychology,Max-Planck Institute for Human Cognitive and Brain Sciences, Leipzig, Germany

*Email:wangf@cbs.mpg.de

Introduction
The Subiculum (Sub) is known as the output layer of the hippocampal formation, and contains boundary vector cells (BVCs), firing for boundaries at specific allocentric directions and distances1, 2. More recently it has been shown that Sub vector cells can exhibits traces that persists for hours after boundary/object removal1 (Fig. 1a).Prior models suggest that such traces can be evoked by place cells (PCs), which index boundary/object presence at encoding2. Vector trace cells (VTCs) mainly occur in the distal Sub (dSub). However, an account of proximo-distal differences remains absent. Here we propose that vector trace cell coding in Sub provides a mismatch signal to update spatial memory representations.

Methods
In our model (Fig. 1b) dSub neurons receive feedforward input from either direct sensory information (BVCs in pSub) or mnemonic information (PCs in CA1). Mismatch between these inputs updates CA1-dSub synapses, with different dSub units having varying updating rates. Following the hypothesized CA1–Sub proximal–distal pathway3, which is implicated in spatial memory specialization, we show how inserted cues affect distal and proximal CA1 (dCA1, pCA1) and their corresponding dSub units. In this model, space-related pCA1 PCs transfer mnemonic information to dSub, while object-related dCA1 PCs exhibit place field drift toward the inserted cue, influencing the probability of synaptic updates between pCA1 and dSub units.
Results
We find that our mismatch-dependent learning model accounts for known VTC properties1, including: (i) the distribution of VTCs along the proximodistal axis, (ii) the percentage of VTCs across different cue types, and (iii) hours-long persistence of vector trace. (iv) By enriching CA1 representations, our model further explains additional empirical findings, including object-centered population coding in CA13. (v) VTCs have longer tuning distances after cue removal.
Discussion
Our model suggests that mismatch detection for updating of associative memory suggests mechanistic explanations for findings in the CA1-Sub pathway, and predicts a function for the Sub in coordinating spatial encoding and memory retrieval. Additionally, it describes the distinctive neural coding for novel objects and familiar contexts and their impacts on memory retrieval. Our work constitutes the first dedicated circuit-level model of computation within the Sub and provides a potential framework to extend the standard model of hippocampal function with a Sub component.

Figure 1. Fig. 1 (a) Experimental Procedure. Rats foraged for food while Sub neurons were recorded. Heatmaps show firing rates as a function of the rat's position (adapted from Poulter et al., 2021). (b) Our model has a perceptual pathway (pSub-dSub) and a memory pathway (CA1-dSub). Arrow widths represent connection strength. dSub units update CA1-dSub weights at varying rates, shown by different colors.
Acknowledgements
AB and FW acknowledge funding from the Max-Planck Society. Additionally, we thank Colin Lever at Durham University for insightful discussions, valuable advice, and access to preliminary data.
References
● Poulter, S., Lee, S. A., Dachtler, J., Wills, T. J., & Lever, C. (2021).Vector trace cells in the subiculum of the hippocampal formation.Nature neuroscience,24(2), 266-275.
● Bicanski, A., & Burgess, N. (2018). A neural-level model of spatial memory and imagery.elife,7, e33752.
● Vandrey, B., Duncan, S., & Ainge, J. A. (2021). Object and object‐memory representations across the proximodistal axis of CA1.Hippocampus,31(8), 881-896.

Speakers

Fei Wang

Andrej Bicanski

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P321: Overcoming the space-clamp effect: reliable recovery of local and effective synaptic conductances of neurons

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P321 Overcoming the space-clamp effect: reliable recovery of local and effective synaptic conductances of neurons

Ziling Wang1,2,3, David McLaughlin*4,5,6,7,8, Douglas Zhou*1,2,3, Songting Li*1,2,3
1School of Mathematical Sciences, Shanghai Jiao Tong University, Shanghai 200240, China
2Institute of Natural Sciences, Shanghai Jiao Tong University, Shanghai 200240, China
3Ministry of Education Key Laboratory of Scientific and Engineering Computing, Shanghai Jiao Tong University, Shanghai 200240, China
4Courant Institute of Mathematical Sciences, New York University, New York, New York 10012
5Center for Neural Science, New York University, New York, New York 10012
6New York University Shanghai, Shanghai 200122, China
7NYU Tandon School of Engineering, New York University, Brooklyn, NY 11201
8Neuroscience Institute of NYU Langone Health, New York University, New York, NY 10016

*Email: david.mclaughlin@nyu.edu, zdz@sjtu.edu.cn, or songting@sjtu.edu.cn
Introduction

To understand the interplay between excitatory (E) and inhibitory (I) inputs in neuronal networks, it is necessary to separate and recover E from I inputs. Somatic recordings are more accessible than those from local dendrites, which poses challenges in recovering input characteristics and distinguishing E from I after dendritic filtering. Somatic voltage clamp methods [1,2] address these issues by assuming an iso-potential neuron. However, as shown in Fig. 1A, this assumption is debated, as the voltage is nonuniform across neurons due to complex morphology [3]. This nonuniform voltage, known collectively as the space clamp effect, leads to inaccurate conductance estimations and even yields erroneous negative conductances [4].

Methods
We study mathematical models of voltage clamping, beginning with an asymptotic analysis of an idealized cable neuron model with realistic time-varying synaptic inputs, and then extending the analysis to simulations of realistic model neurons with varying types, morphologies, and active ion channels. The asymptotic analysis describes in detail the response of the idealized neuron under somatic clamping, and thus captures the discrepancy between the local synaptic conductance on the dendrite, the effective conductances at the soma and the traditional voltage clamp approximation. This discrepancy arises primarily due to the traditional approach’s oversight of the space clamp effect.

Results
With this detailed quantitative understanding of neural response, we refine the traditional method to circumvent the space clamp effect, thus enabling accurate recovery of local and effective conductances from somatic measurements. Specifically, we develop a two-step clamp method that separately recovers the mean and time constants of local conductance on the dendrite when a neuron receives a single synaptic input. Besides, under in-vivo conditions of multiple inputs, we propose an intercept method to extract effective net E and I conductances. Both methods are grounded in perturbation analyses and validated using biologically detailed multi-compartment neuron models with active channels included, as shown in Fig. 1B-1D.

Discussion
Our methods consistently achieve high accuracy in estimating both local and effective conductances through simulations involving various realistic neuron models. Accuracy holds over a broad range of synaptic input strengths, input locations, ionic channels, and receptors. However, two factors can degrade accuracy: large EPSPs and active HCN channels. Large EPSPs, particularly at dendritic tips, require higher-order corrections beyond first-order perturbation theory. Besides, HCN channels also reduce accuracy, but blocking them restores precision. Our approach is robust across various neuron types, as demonstrated in simulations of mPFC fast-spiking neurons, cerebellar Purkinje neurons, and hippocampal pyramidal neurons.

Figure 1. Performance of our method for recovering local and effective conductances in a realistic neocortical layer 5 pyramidal neuron model. (A) Voltage distribution across the pyramidal neuron under somatic voltage clamp condition. (B–D) Our methods perform well in estimating local synaptic conductance features—the mean (B) and time constant (C), as well as the effective conductance at the soma (D).
Acknowledgements
This work was supported by Science and Technology Innovation 2030-Brain Science and Brain-Inspired Intelligence Project (No.2021ZD0200204 D.Z., S.L.); Science and Technology Commission of Shanghai Municipality (No.24JS2810400 D.Z.); National Natural Science Foundation of China (No.12225109, 12071287 D.Z.; 12271361, 12250710674 S.L.) and Student Innovation Center at SJTU (Z.W., D.Z. and S.L.).
References
[1].https://doi.org/10.1038/30735
[2].https://doi.org/10.1016/j.neuron.2011.12.013
[3].https://doi.org/10.1038/nrn2286

[4].https://doi.org/10.1038/nn.2137

Speakers

Ziling Wang

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P322: Optogenetic inhibition of a hippocampal network model

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P322 Optogenetic inhibition of a hippocampal network model

Laila Weyn*1,2, Thomas Tarnaud1,2, Wout Joseph1, Robrecht Raedt2, Emmeric Tanghe1
1WAVES, Department of Information Technology (INTEC), Ghent University/IMEC, Technologiepark 126, 9000 Ghent, Belgium
24BRAIN, Department of Neurology, Institute for Neuroscience, Ghent University, Corneel Heymanslaan 10, 9000 Gent, Belgium
*Email: laila.weyn@ugent.be

Introduction

Optogenetic inhibition of the hippocampus has emerged as a promising approach for suppressing seizures associated with temporal lobe epilepsy (TLE). Given the substantial size of the hippocampus and the inherent challenges of light propagation within the brain, understanding the influence of the volume and nature of the targeted region is crucial. To address these challenges, anin silicoapproach has been developed; allowing systematic exploration of the impact of different target regions on the effectiveness of optogenetic inhibition of seizure like activity in the hippocampus.
Methods
The hippocampal model described by Aussel et al. (2022) was modified and implemented in NEURON [1,2]. A photocurrent described by the double two-state opsin model was added to excitatory neurons of the Dentate Gyrus (DG_E) and Cornu Ammonis 1 (CA1_E) [3]. The impact of hippocampal sclerosis (HS) and mossy fiber sprouting (MFS) modelling [1] on excitability was assessed via an I/O curve of the CA1_E response to DG_E stimulation. Uncontrolled, self-sustaining, high frequency activity was induced in an epileptic network (MFS = 0.9) by reducing the inhibitory component of the EC theta input (see Fig. 1A). The effect of the target region on optogenetic inhibition was studied by varying the number of CA1_E and DG_E cells receiving a light pulse.
Results
The steeper slope of the population response curve suggests that increased MFS correlates with enhanced excitability. For HS, an inverse relationship is observed (Fig. 1B). When 100% of both DG_E and CA1_E regions is illuminated, all activity within the epileptic network is suppressed (Fig 1C.). Reducing the illumination of DG_E allows the network activity to return to theta activity. Notably, illumination of DG_E alone is insufficient to suppress high-frequency firing. These findings indicate that CA1 serves as a better target region for inhibiting hippocampal activity.
Discussion
The results regarding HS and MFS are in line with those observed by Aussel et al. (2022), though a different type of seizure-like activity is generated. Furthermore, the study shows the importance of selecting the appropriate stimulation region to effectively suppress hippocampal seizures. This preliminary investigation explores the capabilities of the network model but further investigation into the generation of seizure-like activity is necessary. Future work will aim for experimental validation of the model generated seizure-like activity and its response to optogenetic inhibition, with the ultimate aim of optimizing stimulation protocols.

Figure 1. A. Healthy and epileptic network response to EC theta current input and optogenetic modulation of CA1 and DG. B. Population response of CA1_E as a function of DG_E activity after stimulation at varying MFS and HS levels. C. Spike count in CA1_E and DG_E populations during optogenetic modulation (t = 1.75:2.25s) of varying amounts of neurons.
Acknowledgements

This work is supported by BOF project SOFTRESET.

References

[1]https://doi.org/10.1007/s10827-022-00829-5
[2]https://doi.org/10.1017/CBO9780511541612

[3]https://doi.org/10.3389/fncom.2021.688331

Speakers

Laila Weyn

Thomas Tarnaud

Wout Joseph

Robrecht Raedt

Emmeric Tanghe

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P323: Modularity and inhibition: the transition from burst-suppression to healthy EEG signals in a microscale model

P327: Deciphering the Dynamics of Memory Encoding and Recall in the Hippocampus Using Information Theory and Graph Theory

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P327 Deciphering the Dynamics of Memory Encoding and Recall in the Hippocampus Using Information Theory and Graph Theory

Jess Yu*1, Hardik Rajpal2, Mary Ann Go1, Simon Schultz1

1Department of Bioengineering and Centre for Neurotechnology, Imperial College London, United Kingdom, SW7 2AZ
2Department of Mathematics and Centre for Complexity Science, Imperial College London, United Kingdom, SW7 2AZ

*Email: jin.yu21@imperial.ac.uk

Introduction
Alzheimer's disease (AD) profoundly impairs spatial navigation, a critical cognitive function dependent on hippocampal processing. While previous studies have documented the deterioration of place cell activity in AD, the mechanisms by which AD disrupts information processing across neural populations remain not fully understood. Traditional analyses focusing on individual neurons fail to capture the collective properties of neural circuits. We hypothesized that AD pathology disrupts not only individual cellular encoding but also the integration and sharing of spatial information across functional neuronal assemblies, leading to compromised spatial navigation.
Methods
We analysed hippocampal CA1 neural recordings from two-photon calcium imaging from AD and wild-type (WT) mice, both young and old, during spatial navigation tasks in familiar environments and novel environments. At the single-cell level, we quantified spatial information using mutual information (MI) between neural spikes and location, and partial information decomposition (PID) [1] for the pair of neurons and location. For population-level analysis, we constructed functional networks using pairwise MI, identified stable functional neuronal assemblies using Markov Stability detection [2], and applied PID to quantify how assemblies collectively encode spatial information through redundancy, synergy, joint mutual information, and redundancy-Synergy Index.
Results
Our analysis revealed multi-scale disruption of spatial information processing in AD. At the single-cell level, AD-Old (ADO) mice showed significantly reduced spatially informative neurons and lower spatial information content. At the assembly level, we uncovered profound deficits in information integration. ADO assemblies showed significantly reduced redundancy and synergy compared to WT-Young controls, indicating impaired information sharing. The Redundancy-Synergy Index revealed a significant shift in the balance between redundant and synergistic processing across neural assemblies.
Discussion
These findings provide novel insights into how AD disrupts neural information processing across multiple scales. The parallel degradation of both cellular encoding and assembly-level information integration suggests a compound effect of AD pathology on spatial navigation circuits. The reduced information sharing between assemblies points to a breakdown in coordinated activity necessary for effective spatial navigation. This multi-scale information-theoretic approach reveals that AD impairs not just individual neural responses but the mechanisms by which neural assemblies integrate spatial information, potentially guiding development of assembly-level therapeutic strategies.

Acknowledgements
This work was supported by the Engineering and Physical Sciences Research Council (EPSRC) through the Physics of Life grant [EP/W024020/1].
References
[1] Williams, P. L., & Beer, R. D. (2010). Nonnegative decomposition of multivariate information. arXiv preprint arXiv:1004.2515. https://doi.org/10.48550/arXiv.1004.2515
[2] Delvenne, J.-C., Yaliraki, S. N., & Barahona, M. (2008). Stability of graph communities across time scales. arXiv preprint arXiv:0812.1811. https://doi.org/10.48550/arXiv.0812.1811

Speakers

Mary Ann Go

Simon R Schultz

Jess Yu

Hardik Rajpal

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P328: Modelling the impacts of Alzheimer’s Disease and Aging on Self-Location and Spatial Memory

P331: Vitrual Brain Inference(VBI): A Toolkit for Probabilistic Inference in Virtual Brain Models

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P331 Vitrual Brain Inference(VBI): A Toolkit for Probabilistic Inference in Virtual Brain Models

Abolfazl Ziaeemehr*¹, Marmaduke Woodman¹, Lia Domide², Spase Petkoski¹, Viktor Jirsa¹, Meysam Hashemi¹

¹ Aix Marseille Univ, INSERM, INS, Institut de Neurosciences des Systèmes, Marseille, France
² Codemart, Cluj-Napoca, Romania*Email: abolfazl.ziaee-mehr@univmail.com

IntroductionUnderstanding brain dynamics requires accurate models that integrate neural activity and neuroimaging data. Virtual brain modeling has emerged as a powerful approach to simulate brain signals based on neurobiological mechanisms. However, solving the inverse problem of inferring brain dynamics from observed neuroimaging data remains a challenge. The Virtual Brain Inference (VBI) [1] toolkit addresses this need by offering a probabilistic framework for parameter estimation in large-scale brain models. VBI combines neural mass modeling with simulation-based inference (SBI) [2] to efficiently estimate generative model parameters and uncover underlying neurophysiological mechanisms.MethodsVBI integrates structural and functional neuroimaging data to build personalized virtual brain models. The toolkit supports various neural mass models, including Wilson-Cowan, Montbrió, Jansen-Rit, Stuart-Landau, Wong-Wang, and Epileptor. Using GPU-accelerated simulations, VBI extracts key statistical features such as functional connectivity (FC), functional connectivity dynamics (FCD), and power spectral density (PSD). Deep neural density estimators, such as Masked Autoregressive Flows (MAFs) and Neural Spline Flows (NSFs), are trained to approximate posterior distributions. This SBI approach allows efficient inference of neural parameters without reliance on traditional sampling-based methods.
ResultsWe demonstrate VBI’s capability by applying it to simulated and real neuroimaging datasets. The probabilistic inference framework accurately reconstructs neural parameters and identifies inter-individual variability in brain dynamics. Compared to traditional methods like Markov Chain Monte Carlo (MCMC) [3] and Approximate Bayesian Computation (ABC), VBI achieves superior scalability and efficiency. Performance evaluations highlight its robustness across different brain models and noise conditions. The ability to generate personalized inferences makes VBI a valuable tool for both research and clinical applications [4], aiding in the study of neurological disorders and cognitive function. Look at Fig.1 for the workflow.
DiscussionVBI provides an efficient and scalable solution for inferring neural parameters from brain signals, addressing a critical gap in computational neuroscience. By leveraging SBI and deep learning, VBI enhances the interpretability and applicability of virtual brain models. This open-source toolkit offers researchers a flexible platform for modeling, simulation, and inference, fostering advancements in neuroscience and neuroimaging research.

Figure 1. Overview of the VBI workflow: (A) A personalized connectome is constructed using diffusion tensor imaging and a brain parcellation atlas. (B) This serves as the foundation for building a virtual brain model, with control parameters sampled from a prior distribution. (C) VBI simulates time series data corresponding to neuroimaging recordings. (D) Summary statistics, including functional connectivit
Acknowledgements
This research was funded by the EU’s Horizon 2020 Programme under Grant Agreements No. 101147319 (EBRAINS 2.0), No. 101137289 (Virtual Brain Twin), No. 101057429 (environMENTAL), and ANR grant ANR-22-PESN-0012 (France 2030). We acknowledge Fenix Infrastructure resources, partially funded by the EU’s Horizon 2020 through the ICEI project (Grant No. 800858).

References
1.https://doi.org/10.1101/2025.01.21.633922
2. https://doi.org/10.1073/pnas.1912789117
3.https://doi.org/10.3150/16-BEJ810
4.https://10.1088/2632-2153/ad6230

Speakers

Marmaduke Woodman

Spase Petkoski

Viktor Jirsa

Meysam Hashemi

Abofazl Ziaeemehr

Lia Domide

Tuesday July 8, 2025 17:00 - 19:00 CEST
Passi Perduti

Posters, Poster session 3

17:00 CEST

P332: Relating Input Resistance and Sodium Conductance

Tuesday July 8, 2025 17:00 - 19:00 CEST

Passi Perduti

P332 Relating Input Resistance and Sodium Conductance

Laura Zittlow*1, Erin Munro Krull1, Lucas Swanson1
1Mathematical Sciences, Ripon College, Ripon, WI, US*E-mail: laurazittlow@gmail.com
Introduction

The sodium conductance density (gNa) determines an axon’s ability to propagate APs. APs do not propagate if the gNa is too low, while they easily do if the gNa is high. Therefore, there is a sodium conductance density threshold (gNaT) [1]. Preliminary results suggest the gNaT for axons with simple morphologies linearly predicts gNaT for axons with more complex morphologies [2, 3]. To address axons with very complex morphologies, we decided to compare gNaT to input resistance (Rin). Rin, defined as the steady-state voltage to injected current ratio, inherently accounts for the axon’s morphology and electrical properties [4].

Methods
We use NEURON simulations [5] to model Rin and AP propagation from an axon collateral to the end of the main axon. We varied the morphology of an extra side branch to see the effect of axon morphology on Rin and gNaT. For each simulation, we find the Rin and gNaT. We evaluate the impact of location for lengths of 0-6𝜆, several side branch morphologies and lengths from 0-6𝜆, and the location and length of sub-branches.
Results
Our simulations show a 1-1 correspondence between Rin and gNaT under specific morphological changes, modeled as a smooth function. Branch location and length affect Rin and gNaT inversely, with their effects stabilizing as the distances and lengths increase. However, when a short side branch connects at the same point as the simulated branch, an abnormality–“bouncing”–occurs. Because shorter side branches are easier to stimulate, the AP can temporarily move into the said branch and thenbounceout. If only one variable (distance or morphology) changes, the error difference is 10-4in gNaT for a given Rin. However, if “bouncing” occurs, then the error difference is on the scale of 10-2.
Discussion
Our results indicate Rinand gNaTrespond monotonically to changes in axonal morphology unless “bouncing” occurs. This suggests Rincould serve as an alternative measure for axonal morphology when predicting gNaT. It offers a computationally efficient method for estimating gNaT. However, “bouncing” disrupts the smooth relationship between Rinand gNaTby making AP propagation more likely. Moving forward, we aim to compare Rin across more complex morphologies. Additionally, we plan to curve-fit the Rin-gNaT relationship and test it against the linear estimation method and realistic axonal morphologies.

Acknowledgements
Thank you to my mentor Dr. Erin Munro Krull and the rest of the Ripon College Mathematical Sciences department for the advice and guidance. Also, thank you to Ripon College's Summer Opportunities for Advanced Research (SOAR) program and the many donors who help fund the program.
References
[1]https://doi.org/10.1152/jn.00933.2011
[2]https://doi.org/10.1186/s12868-018-0467-3
[3]https://doi.org/10.1186/s12868-018-0467-3
[4] Carnevale, N. T., & Hines, M. L. (2006).The NEURON book.Cambridge University Press.
[5] Tuckwell, H. C. (1988).Introduction to theoretical neurobiology: Volume 1. Linear cable theory and dendritic structure.Cambridge University Press.

Speakers

Erin Munro Krull

Reception/party

08:30 CEST

Registration

Wednesday July 9, 2025 08:30 - 19:00 CEST

Registration/organization

09:00 CEST

Workshops

Wednesday July 9, 2025 09:00 - 10:30 CEST

Rooms 4, 5, 6, 9, 101, 103, 202, 203, Hall 1A, Hall 2B, Hall 3B, Belvedere, Onice, Auditorium, or TBA by workshop organizers

Wednesday July 9, 2025 09:00 - 10:30 CEST
Rooms 4, 5, 6, 9, 101, 103, 202, 203, Hall 1A, Hall 2B, Hall 3B, Belvedere, Onice, Auditorium, or TBA by workshop organizers

Workshops

09:00 CEST

BRAIN 2.0: Emerging Research Topics in NeuroAI

Wednesday July 9, 2025 09:00 - 17:30 CEST

Onice Room

One of the goals of the NIH BRAIN (Brain Research Through Advancing Innovative Neurotechnologies) Initiative is to develop theories, models, and methods to understand brain functions and their causal links to behaviors. The modern advances in neuroscience and AI have generated growing interests in NeuroAI, as witnessed by the feedback from the recent NIH BRAIN NeuroAI workshop (Nov. 12-13, 2024). Briefly, NeuroAI is aimed to, first, use AI to understand and improve the brain and behaviors, and second, to develop brain-inspired AI systems for robust, faster and more efficient operations and performances. Motivated by the new wave and developments in NeuroAI, this workshop invites leading experts and new investigators from various research backgrounds to discuss many emerging research topics. The goal of this full-day workshop, in the name of BRAIN 2.0 (BRidging AI and Neuroscience), is to focus on building the bridge between AI and neuroscience, to discuss new research directions and outstanding questions, and to foster team collaborations and open science. Research topics of interest include but not limited to neural transformers and foundation models, new neural network architectures, distributional or meta reinforcement learning, structural reasoning and inference, large language models (LLMs), and digital twins brain. The format of the workshop will consist of both overview-like and research-oriented lecture presentations as well as panel discussions.

Speakers

Zhe Sage Chen

New York University

Michael Halassa

Wednesday July 9, 2025 09:00 - 17:30 CEST
Onice Room

Workshop

09:00 CEST

Brain Digital Twins: from Multiscale Modeling to Precision Medicine

Wednesday July 9, 2025 09:00 - 17:30 CEST

Belvedere room

This workshop will explore how brain digital twins are revolutionizing research into pathological brain conditions and transforming the landscape of precision medicine. Participants will learn how these models work and how they integrate data and tools from different fields, such as molecular neuroscience, network theory and dynamical systems. We will discuss how digital twins can help identify early biomarkers able to characterize pathological states and predict disease progression. Another key topic will be the use of digital twins as in silico environments for testing potential treatments before applying them in clinical scenarios.
Through real-world examples and interactive sessions, we will tackle some of the challenges that come with this innovative approach, such as achieving anatomical precision, handling large datasets, and ensuring ethical use in patient care. The focus will remain on making these cutting-edge tools accessible and impactful, not just for researchers but also for clinicians aiming to deliver more effective, tailored care to their patients.

Speakers

Lorenzo Gaetano Amato

PhD Student, Sant'Anna School of Advanced Study

Alberto Mazzoni

Wednesday July 9, 2025 09:00 - 17:30 CEST
Belvedere room

Workshop

09:00 CEST

Brains and AI

Wednesday July 9, 2025 09:00 - 17:30 CEST

Room 4

Full workshop program

Schedule
9:00-9:30 Fleur Zeldenrust
Heterogeneity, non-linearity and dimensionality: how neuron and network properties shape computation
9:30-10:00 Vassilis Cutsuridis
Synapse strengthening in bistratified cells leads to super memory retrieval in the hippocampus
10:00-10:30 Spyridon Chavlis
Dendrites as nature's blueprint for a more efficient AI
10:30-11:00 Coffee Break
11:00-11:30 Andreas Tolias
Foundation models and digital twins of the brain (online)
11:30-12:00 Robert Legenstein
Spatio-Temporal Processing with Dynamics-enhanced Spiking Neural Networks
12:00-12:30 Max Garagnani
Concept superposition and learning in standard and brain-constrained deep neural networks
12:30-14:00 Lunch
14:00-14:30 Martin Trefzer
Motifs, Modules, and Mutations: Building Brain-like Networks
14:30-15:00 Julian Göltz
From biology to silicon substrates: neural computation with physics
15:00-15:30 Maxim Bazhenov
Do Neural Networks Dream of Electric Sheep?
15:30-16:00 Coffee Break
16:00-16:30 Dhireesha Kudithipudi
Temporal Chunking Enhances Recognition of Implicit Sequential Patterns
16:30-17:00 Thomas Nowotny
Auto-adjoint method for gradient descent in spiking neural networks
17:00-18:00 Questions and Debate
18:00 End of Workshop

Full workshop program

Speakers

Maksim Bazhenov

Thomas Nowotny

Vassilis Cutsuridis

Associate Professor, University of Plymouth

Wednesday July 9, 2025 09:00 - 17:30 CEST
Room 4

Workshop

09:00 CEST

Modeling extracellular potentials: principles, methods, and applications

Wednesday July 9, 2025 09:00 - 17:30 CEST

Room 5

Please visit the dedicated website for full details: https://nicolomeneghetti.github.io/ECP_CNS2025_Wshop/

Simulating large-scale neural activity is essential for understanding brain dynamics and linking in silico models to experimentally measurable signals like LFP, EEG, and MEG. These simulations, ranging from detailed biophysical models to simplified proxies, bridge microscale neural dynamics with meso- and macro-scale recordings, offering powerful tools to interpret data, refine analyses, and explore brain function. Recent advances have demonstrated the clinical and theoretical value of such models, shedding light on oscillations, excitation-inhibition balance, and biomarkers of neurological disorders like epilepsy, Alzheimer's, and Parkinson’s disease. This workshop will cover the latest methodologies, hybrid modeling approaches, and applications of brain signal simulations.By gathering experts across disciplines, it aims to foster collaboration and advance our understanding of brain function and dysfunction.

09:15 – 09:50
Dominik Peter Koller, Berlin Institute of Health (BIH) at Charité – Universitätsmedizin Berlin, Berlin, Germany
Title: "How structural connectivity directs cortical traveling waves and shapes frequency gradients"

09:50 – 10:25
Gaute T. Einevoll, Department of Physics, University of Oslo, Oslo, Norway
Title: "Modeling electric brain signals and stimulation"

10:30 – 11:00
Coffee Break

11:00 – 11:35
Johanna Senk, Institute for Advanced Simulation (IAS-6), Jülich Research Centre, Jülich, Germany
Title: "Large-scale modeling of mesoscopic networks at single-neuron resolution"

11:35 – 12:10
Pablo Martínez Cañada, Research Centre for Information and Communications Technologies (CITIC), University of Granada, Granada
Title: "Inverse Modelling of Field Potentials from Simulations of Spiking Network Models: Applications in Neuroscience Research and Clinical Settings"

12:10 – 12:40
Nicolò Meneghetti, The Biorobotics Institute, Sant’Anna School of Advanced Studies, Pisa, Italy
Title: "From microcircuits to mesoscopic signals: a kernel approach to efficient and interpretable LFP estimation"

12:45 - 14.00
Lunch Break

14:15 – 14:50
Emily Patricia Stephen, Department of Math and Statistics, Boston University, Boston, MA, United States of America
Title: "Connecting biophysical models to empirical power spectra using Filtered Point Processes"

14:50 - 15:25
Madeleine Lowery, School of Electrical and Electronic Engineering, University College Dublin, Dublin, Ireland
Title: "Modelling Neural Activity During Adaptive Deep Brain Stimulation for Parkinson’s Disease"

15:30 - 16:00
Coffee Break

16:00 – 16:35
Meysam Hashemi, Aix Marseille University INSERM, INS, Institute for Systems Neuroscience, Marseille, France
Title: "Principles and Operation of Virtual Brain Twins"

16:35 - 17:10
Katharina Duecker, Brown University and University of Birmingham, USA/UK
Title: "The Human Neocortical Neurosolver as an interactive modeling tool to study the multi-scale mechanisms of human EEG/MEG signals"

Speakers

Nicolò Meneghetti

Post-doctoral fellow, The Biorobotics Institute, Scuola Superiore Sant'Anna Pisa

My name is Nicolò Meneghetti, and I am a postdoctoral fellow at the Computational Neuroengineering Laboratory of the Sant'Anna School of Advanced Studies (Pisa, Italy). My research focuses on computational models of visual processing, as well as the modeling and analysis of extracellular... Read More →

Eleonora Russo

Wednesday July 9, 2025 09:00 - 17:30 CEST
Room 5

Workshop

09:00 CEST

NEW VISTAS IN MULTISCALE BRAIN MODELLING AND APPLICATIONS

Wednesday July 9, 2025 09:00 - 17:30 CEST

Hall 3B

Speakers

Sergio Solinas

Egidio D’Angelo

Rosanna Migliore

Researcher, Istituto di Biofisica - CNR

Computational NeuroscienceEBRAINS-Italy Research Infrastructure for Neuroscience https://ebrains-italy.eu/

Paolo Massobrio

Associate Professor, Univeristy of Genova

Alessia Bonafede

Wednesday July 9, 2025 09:00 - 17:30 CEST
Hall 3B

Workshop

09:00 CEST

Theoretical and experimental approaches towards understanding brain state transitions

Wednesday July 9, 2025 09:00 - 17:30 CEST

Room 6

Speakers

Andre Peterson

The University of Melbourne

Nima Dehghani

MIT

Wednesday July 9, 2025 09:00 - 17:30 CEST
Room 6

Workshop

09:00 CEST

Understanding the Computational Logic of Predictive Processing: A 25-year Perspective

Wednesday July 9, 2025 09:00 - 17:30 CEST

Room 9

Aims and topic
Predictive processes are ubiquitous in the brain and thought to be critical for adaptive behaviours, such as rapid learning and generalisation of tasks and rules. Early works such as the computational vision model proposed by Rao and Ballard (1999) have inspired over two decades of theoretical, computational, and experimental research about predictive neural processing. Stemming from these early works, ongoing investigations provide a rich ecosystem of theory, experiments and computational models that expand beyond the notion of predictive coding. Further, thanks to rapidly developing neural recording technologies, large datasets at multiple scales of granularity and resolution are becoming increasingly available. New computational models enable us to gain a mechanistic understanding of how neural circuits learn to implement and deploy predictive computations. Yet, a full understanding of the underlying computational logic remains fleeting because different aspects are often studied in separate research programs (e.g., layer circuits vs whole-brain neuroimaging), with little cross-pollination. This symposium will look at predictive processing in the context of modern computational neuroscience. Speakers will discuss new theories extrapolating low-dimensional population activity, recent work exploring efficient coding in artificial neural networks and rats' visual cortex, coding hierarchies of prediction errors across brain areas, and computational modelling of behaviour and neural data across species (humans, monkeys, rodents), focusing on high-level, flexible behaviours (hierarchical reasoning, context changes, conceptual knowledge). The topic addressed in this symposium is central to multiple streams of research in computational neuroscience, e.g., perception, decision-making, motor control, and social behaviour. Our aspiration is to stimulate interaction among researchers working in different disciplines and highlight the open questions that will shape future research.

SpeakersMatthias Tsai --Bern University, Switzerland
Rohan Rao --Newcastle University / Oxford University, UK
Erin Rich --New York University, USA
Silvia Maggi --University of Nottingham, UK
Armin Lak --Oxford University, UK
Abhishek Banerjee --Oxford University / Queen Mary University of London, UK
Aurelio Cortese --ATR Institute International, Japan

Schedule
9.00 - 9.05: opening remarks
9.05 - 9.45: Erin Rich,
9.45 - 10.15: Rohan Rao,
10.15 - 10.30: coffee break
10.30 - 11.00: Matthias Tsai
11.00 - 11.45: Aurelio Cortese
11.45 - 12.00: discussion
12.00 - 14.00: Lunch
14.00 - 14.45: Abhishek Banerjee,
14.45 - 15.30: Silvia Maggi
15.30 - 15.45: coffee break
15.45 - 16.30: Armin Lak,
16.30 - 17.00: discussion
17.00 - 17.05: closing remarks

Speakers

Aurelio Cortese

Group leader, ATR Institute International

Aurelio is a group leader at the ATR Institute International in Kyoto, Japan. Aurelio's group is interested in understanding behavioural, computational and neural mechanisms of adaptive decision-making and learning, with an emphasis on metacognition and abstraction. In addition, the... Read More →

Abhishek Banerjee

Professor of Neuroscience, Department of Pharmacology, University of Oxford and Blizard Institute, Queen Mary University of London

Abhi is a Professor of Neuroscience at Barts and Queen Mary University of London and a PI and Wellcome Investigator at the University of Oxford, UK. Abhi's lab is interested in studying neural circuit mechanisms underlying the flexibility of decision-making and how circuit dysfunctions... Read More →