For each data set, Pearson correlation coefficients were calculated for all pairwise comparisons of expressed genes (Supplementary Methods) across all samples. The resulting Pearson correlation matrix was transformed into a matrix of connection strengths (that is, an adjacency matrix) using a power function (connection strength = |correlation|β), which resulted in a weighted network18. WGCNA seeks to identify modules of densely interconnected genes by searching for genes with similar patterns of connection strengths, or high topological overlap18. As a result of the large number of genes analyzed, we carried out an additional step to enrich each network with genes with high topological overlap, which reduced the number of probe sets in each network to 5,549 (CTX), 3,203 (CTX_95), 4,050 (CN) and 4,029 (CB) (Supplementary Methods). For each data set, we used average linkage hierarchical clustering to group genes on the basis of the topological overlap dissimilarity measure (1 − topological overlap) of their network connection strengths18. Using a dynamic tree-cutting algorithm50, we identified 19 modules in CTX, 17 in CTX_95, 23 in CN and 22 in CB (Supplementary Methods). Genes that were not assigned to modules were assigned the color gray.
