The two common methods for filtering imputed data are to combine a minor allele frequency threshold with either the imputed information score >0.3∼0.5 (PROPER_INFO in SNPTEST) [10], [18], [20], [21], [22] or the variance ratio >0.3 (rsq_hat in MACH) [7], [10], [20], [21], [22], [23], [24], [25]. We calculated these two statistics for our data and compared these filters to IQS (Table 3). After filtering by these statistics, the type I error inflation decreases. In the AA sample, IQS also acts as an effective filter and can be cautiously approximated by a combination of MAF and either the imputed information score or the variance ratio (Table S1). Unfortunately, even in the most conservative situation, over three thousand false positives remain. Therefore this is an ineffective approach for filtering poorly-imputed SNPs.
