Our study sample consisted of admixed individuals for which there are two major ancestral founder populations present in highly unequal admixture proportions. Conclusions drawn from this relatively simple admixture scenario may not apply to more complicated scenarios that include samples with more ancestral founder populations and/or more equally distributed admixture proportions. With this caveat in mind, our results lead to two suggestions. First, given the current MACH implementation, we recommend imputation of missing genotypes separately for each panel for situations requiring multiple reference panels (e.g., admixed populations). We achieved the highest imputation yield and coverage by using the HapMap CEU and YRI reference panels separately and then combining the results, rather than combining multiple reference panels prior to imputation. The optimal method of combining separate sets of imputed SNPs is a question for more investigation. The average extent of European ancestry in African American populations ranges from 3.5% to 22.5% [Parra et al., 1998; Parra et al., 2001], but the admixture proportion in a given individual may range from as low as 1.2% to 77.3% [Xu et al., 2007] and
