We constructed gene co-expression networks separately from control individuals and SCZ cases (Supplementary data file 5), since we wished to assess disease-dependent changes in co-expression for modules of interest15. The co-expression network generated from the controls consisted of 35 modules each containing between 30 and 1,900 genes, along with ~3,600 unclustered genes (Supplementary data file 5). Four modules stand out in harboring an excess of differentially expressed genes (Fig. 6A, Supplementary data file 6). Of these, however, only one (M2c) shows association with differential expression (OR = 2.3, P = 1 × 10−13) and multiple prior genetic associations with SCZ; the latter encompasses genes in GWAS loci (FE [fold-enrichment] = 1.36, P = 0.04), rare CNV (FE = 1.52, P = 0.051), and rare nonsynonymous variants (FE = 1.18, P = 2 × 10−4) (Supplementary table 3). Given its apparent relevance to SCZ risk, we tested if the co-expression pattern for M2c was perturbed in SCZ samples relative to controls. We used two categories of network-based preservation statistics: (a) testing whether highly connected nodes in a module remain as highly
