we retained a SNP if there was no already retained SNP on the same chromosome within 1MB. We computed hypergeometric p-values for the enrichment of each pruned set of SNPs overlapping peak calls against the pruned GWAS catalog as the background. We estimated separately for each mark a mapping from a p-value to a false discovery rate across tests for all study and reference epigenome combinations by generating 100 randomized versions of the pruned GWAS catalogs shuffling which SNPs were assigned to which study and computing the average fraction of reference epigenome–study combinations that reached that level of significance (in a continuous mapping of p-values to FDR) using randomized catalogs divided by the number based on the actual GWAS catalog.
