We evaluated gene-specific power of rare variant association studies in the EA and AA samples. We used Fisher's exact test, a robust approach for aggregate testing of rare variation at a locus (27), to determine the power to detect an association for each gene harboring rare causal variants with ORs of 1.5 or 5 in 400 cases and 400 controls (18). In both the EA and AA samples, cases and controls were sampled from 1000 individuals selected to minimize any confounding effects of population stratification (fig. S17), with power calculations assuming a type I error rate of α = 0.001. In each sample, power varies widely across loci, and <5% of genes achieve 80% power even when relatively strong effects (OR = 5) are modeled (Fig. 4A); when causal variants are assumed to have an OR of 1.5, no genes achieve 80% power (fig. S18). Furthermore, although the AA sample has uniformly higher power per gene relative to the EA sample (Fig. 4A), caution is warranted because this is largely a function of our modeling assumptions (18).
