Brain-wide connectome inferences using functional connectivity MultiVariate Pattern Analyses (fc-MVPA)

Current functional Magnetic Resonance Imaging technology is able to resolve billions of individual functional connections characterizing the human connectome. Classical statistical inferential procedures attempting to make valid inferences across this many measures from a reduced set of observations and from a limited number of subjects can be severely underpowered for any but the largest effect sizes. This manuscript discusses fc-MVPA (functional connectivity Multivariate Pattern Analysis), a novel method using multivariate pattern analysis techniques in the context of brain-wide connectome inferences. The theory behind fc-MVPA is presented, and several of its key concepts are illustrated through examples from a publicly available resting state dataset, including an analysis of gender differences across the entire functional connectome. Finally, Monte Carlo simulations are used to demonstrate the validity and sensitivity of this method. In addition to offering powerful whole-brain inferences, fc-MVPA also provides a meaningful characterization of the heterogeneity in functional connectivity across subjects.Author summary: The human connectome comprises billions of functional connections between distant brain areas. In recent years, analyses of functional Magnetic Resonance Imaging (fMRI) data have provided large amounts of information exploring the differences in the human connectome across individuals, developmental trajectories, or mental states. However, scientists’ ability to derive strong conclusions from the analysis of these data are often hindered by the sheer number of connections analyzed, where only connections that show exceptionally large effects are able to stand out against that vast background. This leads to results that tend to overemphasize similarities and mask out differences that are either weaker or distributed across multiple individual connections, potentially misleading conceptual models of the human connectome. This manuscript discusses a novel method for the analysis of the human connectome (functional connectivity Multivariate Pattern Analysis) that addresses these limitations and enables strong conclusions from fMRI data by combining classical statistics with modern pattern analysis techniques. This technique is exemplified using a publicly available database of resting state data to characterize some of the main aspects of the human connectome that differ across individuals, and to identify specific differences in the human connectome across gender.