Also, we compared the yield and associated error rate obtained from two common approaches (“fixed error rate” and the “best call”) for imputation genotype calling. In the fixed error rate approach, we retained discrete genotype calls after ascertaining the threshold of posterior probability required to achieve pre-specified concordance between imputed and genotyped SNPs. In the “best call” approach, the discrete genotype call corresponded to the genotype with the highest posterior probability, regardless of concordance or the posterior probability [Huang et al., 2009]. The result of our analyses showed that the error rates using the best call approach were uniformly greater than the fixed error rate of 0.10 (Table IV). The fixed error rate approach favored a single best reference panel (Table II) whereas the best call approach favored the unweighted mixture reference panel at the cost of a higher error rate (Table IV).
