Leveraging multivariate information for community detection in functional brain networks.

Abstract

Brain functioning relies on specialized systems whose integration enables cognition and behavior. Network science provides tools to model the brain as a set of interconnected brain regions wherein those segregated systems (modules) can be identified by optimizing the weights of pairwise connections within them. However, knowledge alone of these pairwise connections might not suffice: brain dynamics are also engendered by higher-order interactions that simultaneously involve multiple brain areas. Here, we propose a community detection algorithm that accounts for multivariate interactions and finds modules of brain regions whose activity is maximally redundant. We compared redundancy-dominated modules to those identified with conventional methods, uncovering a new organization of the transmodal cortex. Moreover, by identifying a spatial resolution where within-module redundancy and between-module synergy are maximally balanced, we captured a higher-order manifestation of the interplay between segregation and integration of information. Finally, we distinguish brain regions with high and low topological specialization based on their contribution to within- or between-module redundancy, and we observed how redundant modules reconfigure across the lifespan. Altogether, the results show a modular organization of the brain that accounts for higher-order interactions and pave the way for future investigations that might link it to cognition, behavior, or disease.