SBayesR PGS weighting

We generated polygenic scores (PGS) using SBayesR [33] – a Bayesian method that takes GWAS summary statistics as input. This method shrinks SNP effect sizes while still maximising variance explained by “binning” SNPs into a mixture of normally-distributed priors, accounting for linkage disequilibrium. SBayesR has been shown to outperform other PGS methods regardless of the underlying genetic architecture of the trait [33, 36]. SBayesR requires two inputs: 1) GWAS summary statistics from which HapMap3 SNPs with imputation INFO filter > 0.8 were extracted, retaining only those SNPs that passed QC in both AAB and UKB and 2) linkage disequilibrium matrices built using HapMap3 SNPs from a subset of 50,000 unrelated Europeans from the UKB. Additional file 2: Table 1 shows the number of SNPs used in each SBayesR analysis (intersection of HapMap3 SNPs across the GWAS discovery and AAB), and Additional file 2: Table 2 shows the SBayesR output. SBayesR was run with the default inputs: –pi 0.95, 0.02, 0.02, 0.01; gamma 0, 0.01, 0.1, 1; chain-length 10,000; burn-in 2000; out-freq 10, and using the –exclude-mhc flag. For height only, there was an additional step to filter GWAS SNPs with the software package DENTIST [37], to remove inconsistent imputed Z-scores based on the linkage disequilibrium reference matrix and observed GWAS Z-scores, which improved convergence of the SBayesR algorithm.