Supplementary Table 3 shows the results from the prediction analyses. The incremental R2 was calculated as the difference in explained variance when adding the PGI to a regression of the residualized phenotype on the first 10 principal components of the genetic data. In the UKB prediction analyses, we included an additional 10 principal components and 106 genotyping batch dummies. We obtained 95% confidence intervals around the incremental R2’s by bootstrapping with 1000 repetitions. Supplementary Table 3 also shows the predictive power of “public PGIs”, which are PGIs constructed using our Repository pipeline based on the largest publicly available GWAS on the phenotype that does not have sample overlap with the prediction cohort4,5,62–70,72,37,73,80–84,42,49,50,52,55,60,61 (we also use http://www.nealelab.is/uk-biobank/). The details of the input GWAS used for each validation cohort for the construction of the “public PGIs” are in Supplementary Table 13.
