observation mirrors our simulation results showing a constant RA as sample size increases (Supplementary Fig. 8). Fourthly, our predictions of the contribution of LD to the RA of PGS can in principle be inflated in the presence of epistatic interactions between causal variants if they are in strong LD or if causal effect sizes are a function of local LD differences between populations. Lastly, although our simple heuristic strategy to identify candidate causal variants worked well in simulations, we expect the use of standard fine-mapping tools to further improve the efficiency of our method. More specifically, fine-mapping posterior probabilities could be used as weights for candidate causal SNPs, which our current heuristics cannot do. An advantage of our heuristic method is that it utilises whole-genome sequencing data and therefore candidate causal variants that are not present in the GWAS may still be used for inference. In contrast, standard fine-mapping tools are limited by the resolution of GWAS summary statistics, although that resolution can be improved using summary statistics imputation39.
