genomic interval containing one or more eSNPs associated with a gene and having a within-eSNP interval of 500 kb or less. For each such independent block, a single Bayes factor is calculated as the mean of the SNP-level Bayes factors within the block; the SNP-level Bayes factor corresponds to the likelihood of the observed GWAS and eQTL P values under the alternative hypothesis that expression changes in the gene mediate disease risk, relative to the likelihood under the null model where the gene is not related to disease. Bayes factors are multiplied for the independent loci, yielding a single per-gene score. P values for these genic scores are estimated using permuted disease GWAS P values to generate a null distribution of Sherlock Bayes factors across all genes. In this study, we used the eQTL derived from the full cohort of 467 Caucasian-inferred individuals, resulting from the expression-on-SNP regression that included the covariate model with the surrogate variables. The Sherlock method takes as input liberally-defined cis-eQTL associations (P < 10−3) and trans-eQTL associations (P < 10−5). For the trans-eQTL, we used a very strict definition to exclude putatively artifactual associations of SNP and gene expression, requiring, in addition to P <
