The ability to perform PRS analyses directly on imputed data can be particularly useful when the base GWAS and target samples are genotyped on a different platform because then there can be a small fraction of overlapping SNPs. For example, of the 725,459 post–quality control SNPs (see Supplementary Note 1) in the UK Biobank genotype data [10], only 31% (222,956) were found in the GIANT Height and Body Mass Index (BMI) GWAS [20,21]. The use of imputed SNPs increases the number of overlapping SNPs to 2,121,036 SNPs. To assess the gain in power when using imputed vs un-imputed data, we performed PRS analyses on height and BMI using UK Biobank genotyped and imputed data, with GWAS summary statistics provided by the GIANT consortium [20,21]. Age, sex, UK Biobank genotyping batch, UK Biobank assessment centre, and 40 principal components were first regressed out from the phenotype and the standardized residuals were used instead.
