Initially, we imputed on the basis of no SNP-level QC, including all directly typed SNPs, regardless of MAF, call rate and HWE. We also imputed on the basis of only those SNPs that passed stringent QC thresholds (call rate >95% for SNPs with a MAF ≥5% and call rate >99% for SNPs with a MAF <5%, HWE exact P>0.0001, MAF >0.01 and removing all SNPs with GC or TA alleles; Table 1). Although imputation biases can occur due to poor clustering of SNPs with miscalled genotypes in the starting dataset, cluster plot checking is not feasible at the genome-wide scale and therefore, it is not implemented in standard GWAS QC.
