An important aspect of any GWAS analysis is the implementation of a series of rigorous quality control (QC) steps before testing for association. These QC procedures help guard against genotyping error, population stratification, sample duplication and other confounders that can affect the analysis results. QC steps are typically applied at the sample- and SNP-specific level. Sample-level QC includes filtering out samples with low call rates, evidence for different ethnic origin, high heterozygosity, relatedness/duplication, gender discrepancies and genotyping batch effects. SNP-level QC includes filtering out SNPs with low call rates and deviation from Hardy–Weinberg equilibrium (HWE) at pre-determined thresholds. It is generally believed that datasets should be stringently quality controlled (QCed) at the marker level before applying imputation approaches. For this reason, lower MAF SNPs tend to also be excluded, as their accuracy can be hampered by poor clustering properties and incorrect automated genotype calling (at least with currently widely used algorithms). Even though such weight is placed on pre-imputation SNP QC, the effects of applying different criteria and thresholds to the starting dataset have not been investigated thus far. In
