Figure 2 depicts the algorithm that determined which PGIs we constructed. In a preliminary step, we obtained GWAS summary statistics for a comprehensive list of 53 candidate phenotypes (see Supplementary Tables 1 and 2, meta-analysed the summary statistics for each candidate phenotype, and calculated the expected R2 for an out-of-sample regression of each candidate phenotype on a PGI derived from its GWAS summary statistics. We calculated this expected R2 from the GWAS summary statistics (see Methods for details). If it exceeded R2 = 0.01, then we used the meta-analysis output to construct a PGI for the phenotype. We call these the “single-trait PGIs.” For each candidate phenotype, we also identified a list of supplementary phenotypes: any other phenotype whose pairwise genetic correlation with the candidate exceeds 0.6 in absolute value. For each candidate with at least one supplementary phenotype, we then calculated the out-of-sample expected R2 of a PGI derived from a joint analysis of the candidate and supplementary phenotype summary statistics. If the expected R2 exceeded 0.01, then we used the joint-analysis output to construct a “multi-trait PGI” for
