The limiting factor of our intersection-of-SNPs strategy for imputing to a common set of SNPs from different arrays is the degree of overlap in genotyped SNPs present across the arrays to be combined. Using the R2 statistic as a measure of imputation quality, this study demonstrated decreasing imputation quality as the number of overlapping genotyped SNPs decreased. However, this effect did not appear to be as dramatic as might be expected. For example, the overlap between the Illumina 1M and 550v3 arrays on chr.22 was ~7,900 SNPs out of the ~14,000 SNPs on the 1M array (56 % of the original number of SNPs), but the reduction in average imputation quality without any filtering was modest (R2 = 0.91 vs. 0.88 in European American and R2 = 0.89 vs. 0.83 in African American). The intersection strategy remained viable even when 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
