Table 2. The predictive performances (in R ² for linear traits and AUC for binary traits) when all markers are included in PRS in simulations for N = 5000 and 10000 using independent SNPs.

N	h2	Linear traits					Binary traits
		Standard	Tdr	Tweedie	Tweedie*tdr	Standard best p	Standard	Tdr	Tweedie	Tweedie*tdr	Standard best p
5000	0.15	0.012	0.053	0.040	0.047	0.048	0.527	0.542	0.530	0.535	0.546
	0.35	0.051	0.214	0.187	0.200	0.204	0.564	0.620	0.595	0.605	0.608
	0.55	0.118	0.407	0.377	0.397	0.399	0.599	0.705	0.680	0.690	0.691
10000	0.15	0.019	0.085	0.074	0.081	0.081	0.537	0.576	0.559	0.567	0.570
	0.35	0.087	0.275	0.255	0.266	0.274	0.585	0.688	0.671	0.682	0.679
	0.55	0.187	0.461	0.444	0.457	0.468	0.632	0.770	0.757	0.768	0.763

For the columns labelled “Standard”, “Tdr”, “Tweedie” and “Tweedie*tdr”, we first applied LD-clumping with an r² threshold of 0.25 to all SNPs, then PRS was derived using all SNPs that remained. There was no selection of p-value thresholds.

The best predictive performance obtained from optimal p-value thresholds using standard PRS are also shown for comparison (under the column “standard best p”). N denotes the total sample size. For binary traits, an equal number of cases and controls are simulated. Tdr: True discovery rate; h²: total heritability explained. The rest of the simulation results are presented in Supplementary Table 2.