We also investigated the weighted mixture approach to reference panel selection using just MACH based on its superior performance from the above analyses. Using STRUCTURE, the HUFS sample showed admixture proportions of 78–81% YRI and 19–22% CEU. Based on these estimates, we generated 30 reference panels, each consisting of 48 randomly chosen phase II YRI founders and 12 randomly chosen phase II CEU founders. On average, the weighted mixture approach yielded slightly fewer SNPs than the unweighted mixture approach (Table II). This result indicated that matching the reference panel to estimates of admixture proportions did not optimize imputation yield.
