Figure 1

Significant regression coefficients for predicting age and schizophrenia as measured by variable importance in the projection (VIP) for the Riemannian partial least squares (R-PLS) model on the COBRE dataset and the multi-subject dictionary learning (MSDL) atlas with \(K = 2\) latent variables, visualised as connectomes and symmetric matrices. Blue values represent connections that are positively associated with the phenotype, that is, an increase in connectivity between two regions with a blue edge would indicate an increase in the phenotype variable. Conversely, red values are connections that are negatively associated with the phenotype, that is, an increase in connectivity between two regions with a red edge would indicate a decrease in the phenotype variable. (a) Shows the connections that increase with age, (b) shows the connections that decrease with age, and (c) shows the average coefficient values for age between the 17 resting state networks of the MSDL atlas²⁶ (Figs. S2–S7). (d) Shows the connections that increase for patients with schizophrenia, (e) shows the connections that decrease for patients with schizophrenia, and (f) shows the average coefficient values for schizophrenia between the 17 resting state networks of the MSDL atlas²⁶ (Figs. S2–S7). The darker outlined boxes in (c) and (f) show the top \(25\%\) influential regions as measured by the absolute coefficient value within and between each network. The network abbreviations in (c) and (f) are: Aud auditory, Striate striate, DMN default model network, Occ Post occipital posterior, Motor motor network, R V Att right ventral attention network, Basal Basal Ganglia, L V Att left ventral attention network, D Att dorsal attention network, Vis Sec secondary visual cortex, Salience salience network, Temporal temporal network, Language language network, Cereb cerebellum, Dors PCC dorsal posterior cingulate cortex, Cing-Ins cingulate-insula network, Ant IPS anterior intraparietal sulcus.