We constructed a weighted gene coexpression network for each data set on the basis of the pair-wise Pearson correlations for all expressed genes (for a brief glossary of WGCNA terminology, see Supplementary Methods online). The Pearson correlation matrix for each data set was transformed into a matrix of connection strengths using a power function, which resulted in a ‘weighted’ network18. WGCNA seeks to identify modules of coexpressed genes with high topological overlap, a pair-wise measure that describes the similarity of two genes’ coexpression relationships with all other genes in the network18,25. Genes with high topological overlap are therefore highly correlated with the same genes in the network. To test the ability of topological overlap to predict known functional relationships in the human brain, we first asked whether pairs of proteins that physically interact have higher topological overlap in gene coexpression networks than pairs that do not. We obtained a set of 17,540 experimentally validated interacting human protein pairs from EBI (European Bioinformatics Institute)/ IntAct26 and cross-referenced them with expressed genes in CTX, CN and CB. Mean topological overlap was significantly
