Fig. 5: Comparative analysis of GWM-HFN and GM-GM Networks.

A Scatterplot illustrating the relationship between GWM-HFN and GM-GM connection strengths for a representative participant, showing a high correlation (r = 0.743). B, C Heatmaps highlighting edges with the lowest (<5th percentile) and highest (>95th percentile) correlations between the two network types, respectively. D Network-level correlation analysis revealing stronger inter-network correlations (mean r = 0.84) than intra-network correlations (mean r = 0.79) between GWM-HFN and GM-GM connectivity. E Violin plots comparing the magnitude of “true” inter-subject variability between the two frameworks after correcting for intra-subject measurement noise (N). No significant difference was found (Wilcoxon signed-rank test, p = 0.73). F Scatter plot illustrating the moderate correlation of node-wise “true” inter-subject variability between the GWM-HFN (y-axis) and GM-GM (x-axis) frameworks across 90 brain regions. The dashed line represents identity (y = x), and the solid red line is the linear regression fit. G Lateral views of the top 20% (18 out of 90) nodes with the highest “true” inter-subject variability for the GWM-HFN framework (top row) and the GM-GM framework (bottom row). The minimal overlap (Jaccard index = 0.125) highlights the distinct spatial topographies of variability captured by each method. H Line plot of average Jaccard coefficients across sparsity thresholds (0.01–0.50), with shaded regions representing standard deviation, demonstrating consistency across thresholds. I Box plots comparing the area under the curve (AUC) for global topological properties—including clustering coefficient, assortativity, characteristic path length, global efficiency, and small-worldness—across the 0.10–0.34 sparsity range, indicating significant differences between GWM-HFN and GM-GM networks. n = 572 biologically independent participants (SLIM dataset).