P093 A method for generalizing validated single neuron models to arbitrary dendritic tree morphologies
Naining Ge, 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
Single neuron models are vital for probing neuronal excitability, yet their electrophysiological properties remain tightly coupled to individual morphologies as in databases like the Allen Cell Types [1], hindering structure-function studies. Current frameworks, such as evolutionary algorithms linking morphology to electrical parameters [2] and compartment-specific adaptations based on input resistance [3], lack scalability, raising questions about robustness when applied to the variability observed in thousands of neurons.
Methods We introduced a method to adjust single neuron models using morphological features and to validate their generalizability. We tested whether adjusting membrane conductance proportionally to dendritic surface ratios in thousands of single neuron morphologies enables robust generalization of electrophysiological features across morphologies. We validated generalization via two simulation phases: each (1) Allen-fitted model and (2) generalized model adapted to the remaining same-species morphologies. We compared electrophysiological features from Allen-fitted models, simulations (1) and (2) against experimental data. We used an MLP to further refine parameters using morphological features.
Results Total dendritic surface area emerged as a decisive morphological feature that correlates with various experimentally measured electrophysiological features (e.g., rheobase, frequency-intensity slope). Generalization using the method proposed by Arnaudon et al. [3] led to artifactual firing properties in a large subset of the tested morphologies. When we generalized models normalizing total dendritic passive conductance, models showed responses within experimental ranges, demonstrating good biological fidelity. MLP-based prediction reached 15% mean absolute error in the prediction of model parameter sets.
Discussion Our results suggest a promising path towards generalization of validated single neuron models to arbitrary morphologies within a defined electrophysiological cell type. By adapting existing validated models to a broad range of single neuron morphologies, our method offers a framework for large-scale studies of structure-function relationships in neurons and establishes a foundation for optimization of multi-scale neural networks.
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/10.1038/s41467-017-02718-3 [2]https://doi.org/10.1016/j.patter.2023.100855
