The partial least square discriminant analysis (PLSDA) was conducted to define which genes contribute to discriminate between each of the four groups. The plots depict the loadings of each gene: the larger the loading, the better the gene discriminates the study group from the others. Loadings summarise how the genes are related to each other as well as discriminate between the groups: all genes with positive loadings are positive correlated with each other and negatively correlated with genes with negative loadings; colours indicate the group for which the genes have a maximal median value. Panel A (on the three depressed groups only) shows that P2RX7, and, less, CXCL12 and IL-1-beta (all in red), best discriminate TRD vs. the other depressed groups; CCL2, and, less, FKBP5 and MIF (all in green), best discriminate drug-free vs the other depressed groups; and GR, and, less, IL-6 and A2M (all in blue), best discriminate responsive vs. the other depressed groups. Panel B (on the four groups) shows GR (in black) is the gene that best discriminates controls from all the other depressed groups.