We evaluated two practical approaches to minimize the occurrence of low frequency SNPs that are most likely monomorphic in African Americans. First, we used the straightforward approach of imposing an r2hat threshold (e.g., the widely applied r2≥0.3 threshold) on the imputed SNP set, which reduced, but did not eliminate, the occurrence of problematic low frequency SNPs. For instance, 32.7% of the SNPs imputed based on the ALL panel were monomorphic in the YRI+CEU+ASW panel; after imposing the r2hat≥0.3 threshold, 11.0% of the remaining SNPs were monomorphic in the more closely related panel. As an alternative approach, SNPs can be filtered out based on their MAF in reference panel subpopulations (e.g., removing SNPs from the ALL reference panel that are monomorphic in the YRI+CEU+ASW panel) before or after imputation. For pre-imputation filtering, IMPUTE2 offers a filtering option (“-filt_rules_1”) that removes reference SNPs that are monomorphic in the panel of interest (e.g., YRI+CEU+ASW for African Americans), hence leaving fewer input SNP genotypes and speeding the imputation procedure. For postimputation filtering, SNPs are imputed using all available reference SNPs as input and then
