MR-LFADS: A Variational Autoencoder for Disentangling Neural Communication
Multi-Region Latent Factor Analysis via Dynamical Systems (MR-LFADS) is a new sequential variational autoencoder that improves the accurate identification of communication between neural populations. It disentangles inter-regional communication, external inputs, and local dynamics, outperforming existing methods in simulations and demonstrating predictive power on real electrophysiology data for brain-wide circuit perturbations.