Table 1 summarises the number of SNPs with significantly (P<10−6) different allele frequencies between the directly typed and imputed data in the same set of individuals for each of the different QC sets. Correlation plots and R2 values for the comparisons of the QCed and unQCed datasets are presented in Figure 2. The difference between the unQCed (R2=0.993) and QCed data (R2=0.994) was minimal. After post-imputation filtering there were 77 SNPs with significantly different (imputed v. directly typed) allele frequencies in the unQCed data compared with 67 significant SNPs in the QCed data. In an attempt to improve imputation for the small subset of poorly imputed SNPs in the QCed data, we excluded all SNPs with MAF<5% and, subsequently, also SNPs with MAF<10%. We found that eliminating these lower MAF SNPs before imputation had little effect overall. The R2 for the post-imputation QC filtered comparison with the QCed data was virtually identical both when excluding all SNPs with MAF<5% (R2=0.994) and when excluding all SNPs with MAF<10% (R2=0.991).
