A Graph-Based, In-Memory Workflow Library for Brain/MINDS 2.0 – The Japan Digital Brain Project
Carlos Enrique Gutierrez*¹, Henrik Skibbe², Yukako Yamane¹, Kenji Doya¹
¹ Neural Computation Unit, Okinawa Institute of Science and Technology, Okinawa, Japan
² Brain Image Analysis Unit, Riken Center for Brain Science, Wako, Japan
*Email: carlos.gutierrez@oist.jp Introduction
The Brain/MINDS 2.0 Digital Brain Project aims to develop an open, interoperable software platform for digital brain construction. This platform targets the integration of data and leading neuroscience simulation tools such as TVB, NEST, NEURON, BMTK, and others via Python APIs into models of the brain. With the increasing complexity of brain modeling and simulation, the need for platforms capable of integrating heterogeneous datasets and running scalable, high-performance simulations is critical [1,2,3]. Traditional workflow tools like Snakemake and Nextflow, though effective for data pipelines, are limited by their reliance on serialized I/O-based data exchanges, making them less suitability for neural simulation workflows.
Methods
We introduce a graph-based in-memory workflow framework in which diverse brain modeling processes-including parameterization, network building, stimulation, recording, simulation, optimization, and data analysis-are modularized as reusable nodes. These nodes exchange complex neural data through direct memory references. A node-edge graph structure maps both data flow and workflow logic, facilitating transparency and flexibility. Our framework also analyzes data exchange patterns to delineate optimal boundaries, determining when in-memory tasks should be grouped or modularized with file-based I/O for broader compatibility.
Results
Our approach enables the creation of flexible, composable scientific pipelines that support high memory efficiency and reusability for large-scale neural simulations. The framework maintains modular workflow clarity without sacrificing simulation performance. Nodes operating in shared memory facilitate rapid iterative development while enabling scalable extensions. Furthermore, the explicit graph model forms the basis for future enhancements, such as graphical workflow editors and AI-based copilots, which will further broaden accessibility and collaboration.
Discussion
This graph-based, in-memory workflow library addresses a major gap in neuroscience simulation infrastructure by reconciling workflow modularity with high-performance requirements. The system's hybrid capability-integrating in-memory operations with traditional file-based tools like Snakemake and Nextflow-enables distributed execution when needed. This approach encourages the construction of reusable, easily understood workflow components while supporting both collaborative development and large-scale computational neuroscience projects.
This work was supported by Brain/MINDS 2.0 project (AMED), Development of the "digital brain" and related research platforms utilizing mathematical models. 1. https://doi.org/10.3389/fninf.2022.855765
2. https://doi.org/10.1371/journal.pbio.3002158
3. https://doi.org/10.1371/journal.pcbi.1013087
