Fig. 4: The eFC line-graph connectome and model performance of eCPM.
a We used the open-access Gephi (https://gephi.org/) software to visualize edge-centric connectome. The connectome-based plot and icons were automatically generated by inputting the full-length edge-to-edge matrix into this software. To ensure readability, this connectome density has been threshold to 0.1 and was adjusted by using the Fruchterman-Reingold layout. SMN sensorimotor network, DMN default mode network, VIS visual network, VAN ventral attention network, Cont frontoparietal network. Subscripts embedded in abbreviations of networks (i.e., DMNa, DMNb, DMNc, SMNa, and SMNb) indicated the subnetworks within themselves. b It showed the inter-subject correlations between eFC and the cb factor scores, and brain networks parceled by the Yeo-7 network atlas for improving readability. c To show the trained model performance, we provided scatter plots for the correlation between true cb factor scores and predicted ones (z-scored) within the discovery sample (One-sided Permutation test at n = 5000, uncorrected). The Taylor diagram was drawn to comprehensively evaluate model performance by including models that trained from positive eFCs, negative eFCs, and both of them, respectively. d–f We further displayed edge-centric connectome, along with scatter plots and the Taylor diagram to show the model performance in the external validation and generalization in these independent samples (One-sided Permutation test at n = 5000, uncorrected), respectively. Source data are provided as a Source Data file.
