We sought to retain sets that were relevant to the DLPFC brain expression we observed here, as well as address overlap between the 3 databases, using the following strategy. We only retained a gene set in which at least 10% of the genes are expressed in DLPFC (that is, are among the 16,423 genes passing the expression-level threshold. For each set, we filtered out any genes not expressed in DLPFC. We then retained only sets with a final number of genes between 10 and 1,000. For adding the latter two databases, we did not include any set with a Jaccard overlap index > 0.5 to a GO set already included (since in such cases, a substantial portion of the genes were already included in the GO set and the added test would likely be redundant). This procedure yielded 2,902 gene sets in total: 1,938 sets from GO, 824 from Reactome, and 140 gene families.
