RI: Factorial state space models to disentangle multiscale dynamics in cortex-wide activity

Brain-wide recordings now capture activity from thousands of neurons at once, but this activity reflects many underlying processes: sensation, movement, arousal, decision-making, and changes in internal state. Current models tackle these data as if they were generated by a single underlying process, evolving at a single speed, making it difficult to separate overlapping processes that occur simultaneously and unfolding on timescales ranging from milliseconds to minutes. This project develops new mathematical tools that disentangle overlapping brain processes, revealing where each one resides in the brain and how they blend together to support behavior. The project's novelties are models that simultaneously recover many parallel brain processes across a wide range of timescales, together with a principled way to measure when and how those processes interact. The project's broader significance and importance are a testable understanding of how distributed brain activity is coordinated across scales to support perception, decision-making, and natural behavior. The research develops a family of factorial state-space models in which observed neural activity is represented as the sum of multiple latent components evolving in parallel, with distinct spatial footprints, timescales, and dynamical rules. Scalable fitting methods, principled model selection, and built-in measures of uncertainty make these models practical for brain-wide recordings while keeping them interpretable. The models are validated on controlled simulations and benchmarked against current state-of-the-art approaches and then applied to cortex-wide calcium-imaging datasets from behaving mice. The analyses test three central questions in neuroscience: how optimal performance during decision-making relies on stronger routing of information from sensory to motor regions; how fast and slow neural processes combine to produce spontaneous behavior; and how serotonergic neuromodulation triggers slow oscillations that coordinate faster cortical patterns. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.