Regional, but not brain-wide, graph theoretic measures are robustly and reproducibly linked to general cognitive ability

Abstract

General cognitive ability (GCA), also called “general intelligence,” is thought to depend on network properties of the brain, which can be quantified through graph theoretic measures such as small worldness and module degree. An extensive set of studies examined links between GCA and graphical properties of resting state connectomes. However, these studies often involved small samples, applied just a few graph theory measures in each study, and yielded inconsistent results, making it challenging to identify the architectural underpinnings of GCA. Here, we address these limitations by systematically investigating univariate and multivariate relationships between GCA and 17 whole-brain and node-level graph theory measures in individuals from the Adolescent Brain Cognitive Development Study (n = 5937). We demonstrate that whole-brain graph theory measures, including small worldness and global efficiency, fail to exhibit meaningful relationships with GCA. In contrast, multiple node-level graphical measures, especially module degree (within-network connectivity), exhibit strong associations with GCA. We establish the robustness of these results by replicating them in a second large sample, the Human Connectome Project (n = 847), and across a variety of modeling choices. This study provides the most comprehensive and definitive account to date of complex interrelationships between GCA and graphical properties of the brain’s intrinsic functional architecture.

connectome, general cognitive ability, graph theory, multivariate predictive modeling, resting state fMRI