Fig. 2.

Within- and between-subject similarity (WSS and BSS, respectively) for EC and corrFC.A) Matrix of similarity values $S_{\text{EC}}$ between all pairs of sessions from Dataset A1. EC matrices from all sessions are transformed into vectors (see Fig. 1C), from which the PCC is calculated to obtain $S_{\text{EC}}$ for all pairs of sessions (see Eq. (12) in Methods). Here the sessions are grouped by subjects, as indicated by the colored symbols. B) The left panel shows that distributions of WSS (blue) and BSS (red) values for Datasets A1 —corresponding to diagonal and off-diagonal blocks in panel A, respectively— and of BSS (green) for Dataset A2. The right panel shows the corresponding distributions for corrFC. The above error bars represent the means and standard deviations, indicating a smaller overlap between WSS and BSS for EC. C) Visualization of the sessions of Dataset A1 in the space of the first 6 principal components, or PCs (split into the left and right panels), obtained from PCA for EC (top row) and corrFC (bottom row). Each point corresponds to a session and each color to one of the 6 subjects, as in panel A. D) Silhouette coefficients of each session in panel C. Sessions with value close to 1 are well clustered and those close to −1 are poorly clustered; see Eq. (15) in Methods for further details. E) Distribution of the silhouette coefficients for EC (top panel) and corrFC (bottom panel): comparison between the original link space (left) and the PCA space (right, curve in red corresponding to top panel in D). Both Datasets A1 (6 subjects with 6 PCs, in red) and B (30 subjects with 30 PCs, in blue) are represented by the violin plots; see also Figure S3 about the choice for the number of PCs. Note the larger silhouette coefficients for EC than for corrFC.