Data-driven Modeling of Multi-Region Communication Dynamics using Gated Neural ODEs

Advances in multi-region electrophysiology have enabled neuroscientists to collect state-of-the-art datasets comprised of observations of neural dynamics across brain regions. These resources have the potential to help us better understand the dynamics of how populations of neurons communicate with each other during visual decision making. To this end, we introduce Multi-Region Gated Neural ODEs (MR-gnODE), extending gated neural ODEs to model communication between brain regions through separate gated ODE modules. MR-gnODE aims to leverage new multi-region data to accurately identify communication dynamics while maintaining interpretability. We validate this model by recovering communication dynamics between coupled recurrent neural nets (RNNs) and then demonstrate its utility in modeling multi-region communication in the landmark IBL dataset. This work emphasizes the importance of data-driven discovery of brain-wide communication dynamics from emerging large-scale neural datasets.