We evaluated the accuracy of imputed genotypes by comparing allele frequencies at the same SNP between imputed and true, directly typed data. For each QC-imputation iteration, we performed an allele frequency comparison between the actual directly typed and imputed SNPs. Under perfect imputation, we would expect to see alignment with the null hypothesis of no association. We used SNPTEST (http://www.stats.ox.ac.uk/~marchini/ software/gwas/snptest.html)9 to investigate differences between directly typed and imputed genotypes at the same variants within the same samples, taking into account the distribution of genotype probabilities for each individual. For the purposes of our comparison, we used those SNPs that were directly genotyped in OA cases and also present in the HapMap reference samples. Table 1 summarises the number of these SNPs for each QC threshold.
