Coevolutionary balance of resting-state brain networks in autism

TL;DR

This study investigates how coevolutionary balance in resting-state brain networks differs in autistic adults, revealing that preprocessing choices significantly influence the observed network alterations and their potential as biomarkers.

Contribution

It introduces coevolutionary energy as a novel network-level measure to characterize brain organization in autism, highlighting the impact of global signal regression on results.

Findings

ASD shows more negative global coevolutionary energy with GSR

Higher bipolarity observed in ASD without GSR

Machine learning classifies ASD with up to 77.8% accuracy

Abstract

Autism spectrum disorder (ASD) is associated with atypical large-scale brain organization, yet the functional principles underlying these alterations remain incompletely understood. We examined whether coevolutionary balance, a network-level energy measure derived from signed interactions and nodal activity states, captures disruptions in resting-state functional connectivity in autistic adults. Using resting-state fMRI data from ABIDE I with ComBat harmonization to mitigate multi-site batch effects, we constructed whole-brain networks by combining binarized fALFF activity with signed functional correlations and quantified their coevolutionary energy. In the primary analysis with global signal regression (GSR), the ASD group showed significantly more negative global coevolutionary energy (pFDR < 0.002), higher proportions of agreement links, and lower proportions of imbalanced-same links, indicating a systematic redistribution of local motifs rather than a uniform increase in balance. Because GSR can introduce artifactual negative correlations, we repeated all analyses without GSR. In this sensitivity analysis, whole-brain energy and motif differences were attenuated, but bipolarity, a measure of global two-block signed network organization, became the only FDR-significant metric (pFDR = 0.047), with ASD showing higher bipolarity. Intra-network energy differences did not survive FDR correction under either pipeline. Coevolutionary energy showed modest associations with ADI-R and ADOS scores, none of which survived correction across 720 tests. Machine learning classification achieved 77.8% test accuracy (AUC = 0.79) with GSR and 64.7% (AUC = 0.65) without GSR. These findings suggest that coevolutionary balance captures altered signed network organization in ASD, though the specific metric driving group differences depends on preprocessing choices regarding global signal regression.