Whole-genome sequencing will enable characterization of almost all variants in an individual. However, until this becomes affordable in large collections of samples, genotyping arrays, in concert with statistical imputation of untyped alleles, offer a complementary approach to increase power for previously observed alleles. We therefore evaluated the effect on imputation afforded by the larger HapMap 3 resource and also studied how well imputation performs when applied to lower frequency variants and to CNPs.
