Scientists Find New Way to Understand Brain Activity Across Individuals

Researchers developed a framework that groups neurons into latent groups based on their biological profiles and analyzes population dynamics through group-level representations. Because these assignments are anchored in biological properties consistent across subjects, the learned organization can transfer to unseen animals without retraining or neuron-wise correspondence.

Evaluated on whole-brain calcium imaging in larval zebrafish and large-scale electrophysiology in mouse visual cortex, the framework achieved state-of-the-art performance in neural decoding and manifold analysis: it delivered superior decoding of behavior and stimulus labels while better preserving temporal consistency and local geometric structure of latent trajectories on unseen data. The authors report that providing biological features improves baseline performance, but explicitly using these properties to structure the population achieves even stronger results than treating them as auxiliary inputs. Together, they suggest this offers a promising framework for generalizing population dynamics across subjects and recording sessions.