Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster F in Poster Session F: Thursday, August 6, 1:45 – 3:30 pm, Kimmel Center, Shorin & Rosenthal Rooms

Shared neural circuits for brain-wide nonstationary dynamics

Noga Mudrik¹, Ryan Ly², Oliver Ruebel², Adam Shabti Charles³; ¹Allen Institute, ²Lawrence Berkeley National Lab, ³Johns Hopkins University

Presenter: Adam Shabti Charles

Emerging electrophysiology technologies enable large-scale recordings across multiple brain areas, providing an opportunity to study brain-wide dynamics across behaviors and tasks. This endeavor, however, is often complicated by multiple asynchronous recording sessions (repeated measurements with electrodes reinserted) that capture a distinct, unmatched subset of neurons each time. Existing analyses either study each session separately, or overlook the time-varying nature of brain dynamics. We present a method that leverages multi-regional, multi-session data to uncover interpretable whole-brain dynamics that capture the temporal and spatial complexity of neuronal interactions. Rather than framing unmatched sessions as a limitation, we harness them to learn a coherent representation of neuronal dynamics in a low-dimensional latent space shared across sessions, with distinct projections from each session's observations. Each projection is defined by a few neuronal ensembles (sparse clusters of co-active neurons) whose neuronal members are partially and distinctly observed across sessions. Each ensemble exhibits a temporal activity trace, with the traces of all ensembles forming a per-session latent trajectory whose evolution is driven by the ensemble interactions. These ensembles’ interactions evolve non-stationarily over time and vary between sessions, yet they all arise from a shared set of a few core ensemble networks (`circuits’). The circuits' local decomposition, weighted by their activations at each time point, defines the evolving non-stationary interactions between ensembles. The re-usability of these circuits over time and sessions promotes interpretability, while their changing coefficients support temporal variability. We validate our method on synthetic data, and apply it to multi-regional Neuropixels recordings from mice performing a task. We found how multi-regional circuits encode task variables and show distinct activations in correct vs. incorrect trials. We further revealed that evolving circuit activations influence decision-making accuracy, with the Hippocampus supporting early task encoding and the Secondary Motor Cortex (M2) maintaining memory during the delay period.

Topic Area: Methods, Tools, Theory & Neural Coding