We created two predictors in the QIMR cohort by PLINK15, one based on all 87 multiple associated SNPs and the other based on the 49 additionally associated SNPs only, with SNP effects estimated from the joint analysis using the ARIC cohort as a reference sample, and then regressed the observed height phenotypes on the predictors. We performed the same prediction analysis in the ARIC cohort but with SNP effects estimated from the joint analysis using the QIMR cohort as the reference sample, acknowledging that the ARIC cohort is part of the discovery sample of the GIANT meta-analysis. The regression slopes were not significantly different from 1 (Table 2), suggesting that the estimates of joint SNP effects are unbiased, the prediction R2 of all 87 SNPs was ~3.8–4.8%, consistent with the estimate of 4.1% of variance explained in the discovery sample, and the prediction R2 of the 49 additional associated SNPs was ~1.3–1.5%, in line with the estimate of 1.6% of variance explained by these SNPs in the discovery sample. Hence, by performing a prediction analysis in an independent sample, we confirmed that the 49 additional associated variants explain approximately 1.3% of phenotypic variation.
