The unbiased intersection strategy for imputation across arrays uses fewer genotyped SNPs as the basis for imputation compared to the union strategy. We took two strategies to investigate the impact of using smaller numbers of SNPs in the intersection-based strategy on SNP imputation quality. First we evaluated the imputation quality of SNPs for which we had genotype data but were removed from the intersection set for imputation because they were not present in all arrays. This allowed us to compare the true genotypes to the genotypes imputed using the intersection strategy. In European Americans from SAGE, 40 % of the SNPs genotyped on the 1M array were not genotyped on the 550v3 and thus were masked for the intersection-based imputation strategy. For these SNPs, the “best call” imputed genotypes (Shriner et al. 2010) were highly concordant with their directly typed genotype (97.1 % concordance rate), and 99.6 % of the masked genotyped SNPs were imputed at R2 >0.3 (standard threshold for evaluating imputation quality, Li et al. 2010). A comparable analysis of the masked genotyped SNPs in SAGE African Americans resulted in a 94.7 % concordance rate and 97.1 % of the SNPs having R2 >0.3.
