Extended Data Fig. 2. Predictive accuracy for different polygenic prediction methods in simulations using half GWAS sample size as used in the primary simulation.

Sample sizes of 40K European and 10K non-European were used. Results are shown for BridgePRS, PRS-CSx, PRS-CS-mult and PRSice-meta across six simulation scenarios, with and without the causal variants included in the model for African and East Asian ancestry samples. a SNP heritability ${{\mathrm{h}}^2}_{\mathrm{SNP}}=0.25$ and b SNP heritability ${{\mathrm{h}}^2}_{\mathrm{SNP}}=0.5$, ten simulated phenotypes per scenario. Under each set of analyses the proportion of causal variants and the relative power of the data used is shown, measured by nh²/m up to proportionality, where n is the GWAS sample size, h² heritability and m the number of causal variants. The central rectangular boxes show the interquartile range, horizontal lines inside the boxes show the median, whiskers extend to the most extreme results and points show results for each of the 10 simulated phenotypes.