and HumanOmni1_Quad_v1-0_B beadchips, respectively. The association analysis was performed using unconditional logistic regression models implemented in PLINK (30). Diagnosis and alleles each served as the dependent and independent variables, respectively, with sex, age and the first 10 principal components (PCs) of ancestries as covariates. These PCs were estimated from all autosomal SNPs that had a call rate >95%, using principal component analysis (PCA) implemented in the program EIGENSTRAT (31). These SNPs were stringently cleaned using the QC criteria as we previously published (25) before PCA. Each individual received scores on each principal component. Because related subjects, non-EA and non-AA subjects, and any population group outliers have been excluded during data cleaning, their effects on PCA were removed. Although different sets of autosomal SNPs were used to estimate these PCs in different datasets, the first 10 PCs all explained >95% of variance of the samples in different studies.
