including several arrays; imputation based on ~4,000 overlapping genotyped SNPs across the Illumina 1M, Omni1-Quad, 660W, and 550v3 (~30 % of the ~14,000 on the 1M array for chr.22) showed an average R2 = 0.79 across the MAF spectrum for European Americans and R2 = 0.68 for African Americans. Because of the differing SNP selection strategies, the overlap between the Illumina 1M and Affymetrix 6.0 was low (~3,200 SNPs), resulting in somewhat poorer imputation quality: R2 = 0.73 in European Americans and R2 = 0.61 in African Americans. A related consideration may be the number and character of SNPs being imputed. In this study’s examples, the number of genotyped SNPs changes as one adds arrays to the intersection, but the number of imputed SNPs remains the same. In parallel, it may be that keeping the number of genotyped SNPs the same but increasing the number of imputed SNPs will reduce imputation accuracy as the genotyped SNPs will likely have weaker correlations with the larger set of imputed SNPs and their characteristics change. We have observed this in another study testing differences in imputation performance among African Americans by reference panels and imputation software (Hancock et al. 2012). In that study,
