Accurate Identification of Communication Between Multiple Interacting Neural Populations

Neural recording technologies now enable simultaneous recording of population activity across multiple brain regions, motivating the development of data-driven models of communication between recorded brain regions. Existing models can struggle to disentangle communication from the effects of unrecorded regions and local neural population dynamics. Here, we introduce Multi-Region Latent Factor Analysis via Dynamical Systems (MR-LFADS), a sequential variational autoencoder composed of region-specific recurrent networks. MR-LFADS features structured information bottlenecks, data-constrained communication, and unsupervised inference of unobserved inputs--features that specifically support disentangling of inter-regional communication, inputs from unobserved regions, and local population dynamics. MR-LFADS outperforms existing approaches at identifying communication across dozens of simulations of task-trained multi-region networks. Applied to large-scale electrophysiology, MR-LFADS predicts brain-wide effects of circuit perturbations that were not seen during model fitting. These validations on synthetic and real neural data suggest that MR-LFADS could serve as a powerful tool for uncovering the principles of brain-wide information processing.

Brain function relies on different parts of the brain working together to process our senses, generate our perceptions and thoughts, and drive our bodies into action. New technologies are allowing brain scientists to monitor the activity of large populations of individual neurons simultaneously across many brain regions. These measurements enable scientists to ask questions about how the activity in one region affects another--that is, how brain regions actually communicate. Our research introduces a new machine learning technique that uses multi-region neural activity data to infer the direction and content of communication between brain regions. Unlike existing approaches, our technique explains the neural activity in each measured brain region in terms of communication from other measured brain regions, influences from unmeasured brain regions, and how each brain region internally processes information over time. When applied to simulated brain networks designed to reflect challenging scenarios for studying communication, our technique accurately identified who was communicating with whom, and what signals were being communicated. We then applied the technique to real brain data and showed that it could predict the brain-wide effects of disrupting one region--effects our model had never seen before. Taken together, this work provides a powerful new tool for studying how different parts of the brain work together and may provide insight into developing treatments for brain injuries and disorders.