Within the three compared reference panels (i.e., CEU, YRI and unweighted CEU+YRI), imputation yield was highest for BEAGLE using the unweighted phase II CEU+YRI reference panel (Table II). In contrast, imputation yield was highest for MACH using the phase II YRI reference panel. Notably, imputation yield was higher for MACH than for BEAGLE irrespective of the tested reference panel. PLINK failed to achieve an error rate of 10% for all tested reference panels and so yielded no successfully imputed SNPs. Despite the remarkable success of the YRI as a reference panel for this African American sample, it is important to point out that it cannot serve as a reference for those SNPs (n = 6,782) present in the phase II CEU reference panel but not in the phase II YRI reference panel (Fig. 1). Interestingly, when the phase II CEU reference panel was used as the sole reference panel, the lowest error rate that MACH could achieve for this specific set of SNPs was 22.3%; this observation suggests that high quality imputation may not be achievable for SNPs segregating only in an ancestral population making a minor contribution to an admixed population.
