After correcting for the majority of confounding effects on gene association scores, we next combined gene scores at the level of gene sets. We developed an approach similar to GSEA that tests whether predefined sets of functionally related genes are enriched for genes associated with a given complex disease or phenotype, more than would be expected by chance (Figure 1D). Specifically, the GSEA algorithm in MAGENTA tests for over-representation of genes in a given gene set above a predetermined gene score rank cutoff. The enrichment is evaluated against a null distribution of gene sets of identical set size that are randomly sampled from the genome multiple times (see Materials and Methods for details). The 95th percentile of all gene scores for a given GWA study or meta-analysis was used here as the enrichment cutoff (see Figure S5 for cutoff choice). Since subsets of genes in biological pathways are often physically proximal in the genome [25], for each gene set, we removed all but one gene from each subset of genes assigned the same best SNP, to prevent inflation of an enrichment signal due to positional clustering of genes (assuming one gene per associated variant).
