Fig. 4: Behavioral significance across methods for fMRI resting-state data.

a Percentage of brain cognition covariance explained by significant multivariate correlation components (permutation testing with p < 0.05, 1000 permutations) obtained from Partial Least Square Correlation (PLSC) analyses between the whole-brain functional connections extracted from the different methods and 10 cognitive scores. PLSC components were independently assessed for each method. On a whole-brain level, no differences are noticeable. b We repeat the same PLSC analysis considering only the functional connections extracted from the different methods within specific functional networks (i.e., all interactions between nodes of the same network). Remarkably, higher-order methods perform better in terms of covariance explained, reaching 80% in somatosensory areas. c For each method and focusing on connections within functional networks, we report the cognitive saliences representing the cognitive domains contributing the most to the brain-cognition multivariate correlation patterns. Colored squares represent cognitive domain weights corresponding to the significant PLSC components. Interestingly, methods beyond functional connectivity tend to explain different cognitive dimensions across the various functional networks. Notice that to obtain a more robust estimate of the covariance explained, we consider a bootstrap procedure randomly sampling 80 subjects from the total of 100 and repeating the PLSC analysis 100 times (n = 100). The box plots show the median and interquartile range (IQR), with whiskers extending to Q1-1.5IQR and Q3+1.5IQR. Caps mark the ends of the whiskers. Individual data points, including outliers, are shown using scatter plots. The functional networks analyzed correspond to: visual (VIS), somatomotor (SM), dorsal attention (DA), ventral attention (VA), limbic (L), frontoparietal (FP), default mode network (DMN), and subcortical (SC).