Decomposing the modulation of interactions between neuronal populations

Abstract

Identifying, through analysis of large-scale neural recordings, subpopulations of neurons interacting across brain areas is key for understanding across-brain communication with cellular resolution. The communication subspace (CS) framework addresses this need by capturing additive interactions between subspaces of source and target neural populations. However, no current method identifies interpretable components of how a measured, multivariate modulator - such as behavioral state or the activity of another population - reshapes these interactions. Here we extend the CS framework by using a bilinear perturbation of reduced-rank regression to parameterize modulation as a low-rank tensor. We derive MICs - multiplicative interaction channels - a method to identify triplets made of a source and a target activity pattern and a modulator of their interaction. We develop a hierarchical fitting pipeline and provide a closed-form decomposition that quantifies whether modulation reshapes the modulator-averaged baseline interaction, recruits private dimensions of one population, or opens new interactions. In simulations, MICs reliably recovers the presence and geometry of ground-truth modulation even in the high-dimensional, low-sample regime. Applied to a benchmark macaque dataset, MICs recovers the geometry and functional form of the stimulus modulation of interactions between primary and secondary visual cortices. Applied to simultaneous calcium imaging of prefrontal axons and interneurons in mouse visual cortex, MICs reveals that behavioral state asymmetrically modulates top-down interactions, reconfiguring the prefrontal projection patterns that interact with a stable set of visual interneurons. MICs provides a data-efficient, interpretable account of how measured modulators reshape the interactions between neural populations. Code is available at https://github.com/mcelotto/multiplicative_interaction_channels.