Accurate Identification of Communication Between Multiple Interacting Neural Populations

Area of Science:

• Neuroscience
• Computational Neuroscience
• Systems Neuroscience

Background:

• Advanced neural recording technologies allow simultaneous monitoring of population activity across multiple brain regions.
• Existing data-driven models often fail to accurately distinguish sources influencing neural populations, hindering the study of inter-regional communication.

Purpose of the Study:

• To introduce a novel computational framework, Multi-Region Latent Factor Analysis via Dynamical Systems (MR-LFADS), for disentangling neural communication patterns.
• To improve the accuracy of modeling brain-wide information processing by separating inter-regional communication, external inputs, and local dynamics.

Main Methods:

• Developed MR-LFADS, a sequential variational autoencoder model.
• Utilized dynamical systems principles to model neural population activity.
• Validated the model using extensive simulations of task-trained multi-region neural networks.

Main Results:

• MR-LFADS demonstrated superior performance compared to existing methods in identifying communication across simulated neural networks.
• The model successfully predicted brain-wide effects of circuit perturbations on large-scale electrophysiology data not used during training.

Conclusions:

• MR-LFADS offers a robust approach for analyzing complex neural population dynamics and inter-regional communication.
• The model serves as a valuable tool for uncovering fundamental principles of brain-wide information processing using real and synthetic neural data.