The relationship between allele frequency differences between CEU and YRI and power.

We plot the histogram of all SNPs in the 1000 Genomes Project data as a function the allele frequency difference between CEU and YRI (excluding SNPs monomorphic in both CEU and YRI). The histogram is colour-coded by the estimated change in power by performing phase 2 in YRI instead of CEU, assuming a total sample size of 20,000 (10,000 in CEU in phase 1, and 10,000 in YRI in phase 2) and a GRR of 1.2. Allele frequency differences from +15% to +40% in YRI result in a positive gain in power (in red), which is compensated by SNPs that are common in CEU (in blue). We divide the histogram into 4 categories: (1) SNPs with at least 80% power in both scenarios (CEU in phase 2 or YRI in phase 2) (65.6% of all SNPs considered), (2) SNPs with at least 80% power to detect an association in the European GWAS (CEU in phase 2) (6.5% of all SNPs considered), (3) SNPs with at least 80% power in the African GWAS (YRI in phase 2) (9.3%), and (4) SNPs that do not reach 80% in either of these two scenarios (18.6%). As alleles of higher frequency in CEU are mostly saturated for power, including additional European samples in GWAS will only marginally increase power, whereas alleles of lower frequency in CEU may substantially benefit in terms of power from elevated frequencies in African populations.