We further investigated the optimal strategy for imputing missing genotypes for an admixed sample such as the HUFS. Of the 3,121 successfully imputed SNPs using the phase II CEU reference panel, 394 were present only in the phase II CEU reference panel and 2,727 were present in both phase II CEU and YRI reference panels. Similarly, of the 13,927 successfully imputed SNPs using the phase II YRI reference panel, 3,688 were present only in the phase II YRI reference panel and 10,239 were present in both phase II CEU and YRI reference panels. Taken together, we were able to impute 394 SNPs present only in the phase II CEU reference panel, 3,688 SNPs present only in the phase II YRI reference panel, and 10,760 SNPs present in both phase II CEU and YRI reference panels, for a total yield of 14,842 SNPs. This result clearly demonstrates that, at least for our admixed sample of African Americans, the optimal strategy in terms of maximizing imputation yield is to impute missing genotypes separately from the ancestral reference panels and then combine the results.
