LDSC computes an LD score by summarizing the correlations of a given SNP with all neighboring SNPs within 100 kb flanks. Then, the GWAS test statistic χ2 is regressed against the LD score, of which the slope is rescaled into an estimate of SNP-h2, explained by all SNPs included in the LD score. Based on Sg,c, we binned genes in specificity deciles and used LDSC to test for enrichment of SNP-h2 in the top 10% most associated genes. MAGMA aggregates χ2 association statistics within a 10 kb upstream and 1.5 kb downstream window into a gene-level P value (Supplementary Methods). To compute gene-level P values, Brown’s method [26] was updated into Imhof’s approach [27] in version 1.08 [28]. DEPICT maps genes to loci by first selecting all significant SNPs and preserving lead SNPs as the most significant SNP out of possible SNP-pairs in LD (r2 > 0.1) and/or within < 1 Mb distance of each other (Supplementary Methods). Then, boundaries of the most distal SNP on either side around lead SNPs (r2 > 0.5) were used as criteria to list
