Fig. 1.

Comparison of prediction performance in simulation studies with non-overlapping annotation groups (Settings 1–5), measured by AUC (a) and $R^2$ (b). A total of $M=125,000$ SNPs were sampled from the UK Biobank data with 1 KG as LD reference panel. Genetic effects were generated using a mixture of point-Normal models with total heritability fixed at 0.7. GWAS results from $N_{\mathit{sumstat}}=50,000$ simulated individuals were used as summary statistics. Prediction accuracy was evaluated in a test sample of $N_{\mathit{test}}=24,000$ simulated individuals with tenfold cross-validation