To evaluate GoShifter’s performance in prioritizing annotations that identify causal variants, we would ideally use a set of trait-associated loci in which the causal variants and relevant driving genomic annotations are known. However, causal variants are known for only a handful of complex-trait loci. Therefore, to capture a wide range of possible models of causal variation, we simulated 1,000 sets of 1,416 SNPs by tagging functional SNPs selected from seven distinct functional genomic annotations: DHSs, promoters, 5′ UTRs, nonsynonymous SNPs in exons, 3′ UTRs, introns, and intergenic regions (Figure S2). We then tested these SNP sets for enrichment in DHSs. Pre-defining functional SNPs on the basis of specific driving annotations allowed us to assess the ability of a method to identify true enrichment and reject spurious overlap (Figure S3). An appropriate strategy should detect high DHS enrichment in sets designed to tag functional variants in regulatory regions (DHSs, promoters, and 5′ UTRs), modest enrichment in nonsynonymous variants in exons and 3′ UTRs (which colocalize with DHSs),16,17 and no enrichment at introns or intergenic regions (Material and Methods).