We also used a weighted gene coexpression network analysis (WGCNA) approach (41) to construct coexpression networks and extract gene modules for each tissue (Fig. 6B). The clustering of coexpressed genes into modules allows us to identify active biological processes across tissues (Fig. 6C). Modules enriched for common Gene Ontology (GO) biological processes were observed in all nine tissues (e.g., cell cycle, protein transport), while other biological processes were only seen in specific tissues (e.g., fatty-acyl-CoA metabolic process enrichment seen only in adipose, Benjamini-Hochberg corrected P = 2.6 × 10−7) (14). In addition to functional annotation, the identified coexpression networks also enabled us to search for potential transcriptional regulators of these modules (Fig. 6D). We found enrichment of transcription factor binding in promoter regions of genes in the same module using ENCODE chromatin immunoprecipitation (ChIP)–seq data (42), suggesting that some modules could be regulated by a large number of transcription factors (fig. S30A). Finally, we compared coexpression modules learned in each tissue individually to those learned in other tissues, based on common gene membership and correlation of first principal components of
