We used an empirical GWAS dataset to assess the effect of QC on imputation outcome. We focused on chromosome 22, (n=9038 directly typed SNPs) from 3177 osteoarthritis (OA) cases from the United Kingdom, typed on the Illumina 610k quad chip (Illumina) as part of the arcOGEN consortium GWAS (manuscript submitted). Chromosome 22 is representative of the genome in terms of the proportion of directly typed to imputed SNPs. All samples included in our analysis had passed standard sample level QC (based on call rate, heterozygosity, relatedness, ethnicity and gender discrepancies). We imputed genotypes at variants on the basis of HapMap phase II release 22 CEU data (n=33 815 SNPs on chr22) using IMPUTE v1 (https://mathgen.stats.ox.ac.uk/impute/impute.html).8 We performed each imputation in duplicate, with and without the IMPUTE v1 predict genotyped SNPs flag, which resulted in one set of imputed data containing the original genotypes and in the other imputed genotypes. To assess the effect of varying levels of QC, we carried out several rounds of imputation, using differently QCed OA SNP data as the starting point.
