When comparing directly typed with imputed allele frequencies at the same variant in the same individuals, we arbitrarily considered P<10−6 as significantly different. We calculated the correlation between imputed and directly typed MAF, using the expected counts to allow for genotype-associated probabilities. We also applied a series of post-imputation QC steps in order to eliminate unreliably imputed SNPs, aiming to filter out as many of these SNPs as possible while retaining a good proportion of nonsignificant SNPs. We compared two alternative methods for post-imputation QC filtering, first, the IMPUTE-info score, which is associated with the imputed allele frequency estimate which ranges from 1, indicating high confidence, to 0 suggesting decreased confidence, and second, the freq-add-proper-info score provided by SNPTEST, a relative statistical score ranging from 0 to 1, representing no information to complete information, respectively. The SNPTEST freq-add-proper-info score has been shown to be highly correlated with the IMPUTE-info score under the additive model.10 In both scenarios, we also filtered out SNPs with MAF <5%. Figure 1 illustrates the effects of altering post-imputation QC filters on the QCed data. On
