Figure 3 shows the predictive power of PRSice-2 when compared to lassosum and LDpred for quantitative traits with heritability of 0.2, base sample size of 50,000, and target sample size of 10,000 (see Supplementary Fig. 2 for comparisons across all settings). Consistent with previous findings [15,24, 25], PRSice-2 has comparable predictive power to lassosum and LDpred, typically generating PRSs with predictive power higher than those of LDpred but not as high as lassosum. However, these results are inherently dependent on our modelling assumptions. For example, in our simulation, effect sizes and residual effects are assumed to have a Gaussian distribution and all “causal” SNPs are included in the dataset. Thus, we provide these results only as an approximate guide to performance in settings that match our assumptions. We provide our simulation code [26] for others to inspect and repeat our analyses.
