There are two situations in which imputation is avoided[18]: (1) SNPs with low minor allele frequency and (2) cases and controls genotyped on different platforms. The statistics previously used for measuring the accuracy of imputation are inadequate for evaluating the quality of imputation due to their dependence on marginal SNP frequency. Specifically, imputation accuracy, a measure of the concordance rate between the imputed and observed genotypes for each SNP, dramatically over-estimates reliability when minor allele frequencies are low and does not address the inflation of false positive rates arising from imputation error due to random agreement. We developed IQS to more precisely estimate imputation error, effectively filtering imputation error in these two problematic situations. We showed that IQS is a more appropriate measure to evaluate imputation reliability because it adjusts for “chance” agreement, and filtering by IQS eliminates the inflation of the false positive rate arising from imputation error.
