To account for the effect of population structure on genetic association in these African-American samples, principal components analysis was computed on cohorts 1 and 2 with GWAS data by using all SNPs that passed quality control standards and after exclusion of regions of high LD and inversions. The first principal component (PC1) explained the largest proportion of genetic variation (22%). DNA samples from all cohorts, as well as 44 Yoruba Nigerians and 39 European Americans were genotyped for 77 ancestry informative markers. The African to European ancestral proportion of each African-American sample was then estimated using the Expectation-Maximization algorithm implemented in the program frequentist estimation of individual ancestry proportion (FRAPPE) under a two-population model.39 PC1 was highly correlated with ancestry informative markers (r2 = −0.87), suggesting that PC1 largely reflected the ancestry= proportions, and was used as a covariate for association analysis of cohorts 1 and 2. The mean (±s.d.) African ancestry proportions estimated by frequentist estimation of FRAPPE in cohorts 1 to 6 were 0.77±0.12, 0.78±0.12, 0.77±0.12, 0.78±0.13, 0.76±0.11 and 0.69±0.14, respectively.
