In our Scenario B dataset, we demonstrated that an expanded reference panel containing thousands of chromosomes can greatly improve accuracy over what is possible based on the HapMap alone, although these gains are limited to the subset of HapMap SNPs that are included on multiple genotyping chips. This finding is consistent with the conclusions of the recent BEAGLE paper [13]. IMPUTE v2 was consistently among the most accurate methods we considered. For example, IMPUTE v2 attained best-guess error rates that were 15–20% lower than those of its closest competitor (BEAGLE) in a realistic representation of Scenario B.
