Fig. 1: Specific description of method for group identification.

Groups were uncovered in six steps, as detailed in Supplemental Information: (1) Preprocessing datasets by flattening matrices (Fig. S6); (2) Identifying optimal functional connectivity sets by multilevel NMF factorization: (2.1) Mathematical description of the NMF (Figs. S5-S7); (2.2) Decomposing the data into a multilevel family of sub-matrices (Fig. S7); (3) Dissecting factors in biclusters, including (3.1) Dissecting NMF k factors into sub-matrices or biclusters which are interpreted as fMRI sets (Fig. S7), and (3.2) Learning the W and H matrices of FNMF; (4) Selecting biclusters: Multi-view and optimally assembling the families of sub-matrices (Fig. S10); (5) Statistical analysis of biclusters (Figs. S2, S9); (6) Graph and matrix representations of biclusters (Fig. S8), including (6.1) Displaying biclusters extracted after factorization (Figs. S2, S9), (6.2) Displaying and decoding TBSS biclusters and transforming them back to native space (Fig. 2, S1, S6), and (6.3) Displaying and decoding fMRI biclusters and transforming them back to native space (Fig. 3, S3, S6).