. 2017 Feb 1;7:41262. doi: 10.1038/srep41262

Table 3. Predictive performances (prediction R ² in %) of the standard PRS and four other PRS schemes in simulations using real genotype data (a mixture small and large effects simulated).

% casual	h²	Type	Standard	Tdr	Tweedie	Tweedie*tdr	LDpred
0.1%	10%	max	0.700	0.962	0.951	0.925	0.705
0.1%	10%	all SNPs	0.044	0.644	0.165	0.903	0.055
0.1%	20%	max	2.497	2.712	2.669	2.922	3.192
0.1%	20%	all SNPs	0.072	2.025	0.178	2.909	0.041
0.1%	30%	max	6.622	6.800	6.627	7.079	7.376
0.1%	30%	all SNPs	0.411	4.342	2.852	7.079	0.489
0.1%	40%	max	7.327	8.289	8.522	8.846	7.743
0.1%	40%	all SNPs	0.265	5.048	2.634	8.790	0.817
0.25%	10%	max	0.921	1.154	1.269	1.328	0.986
0.25%	10%	all SNPs	0.075	0.864	0.237	1.327	0.162
0.25%	20%	max	2.381	2.446	2.342	2.432	0.913
0.25%	20%	all SNPs	0.038	2.099	0.320	2.416	0.038
0.25%	30%	max	2.015	2.469	2.328	2.571	2.805
0.25%	30%	all SNPs	0.730	1.899	1.268	2.571	0.653
0.25%	40%	max	4.762	5.328	5.218	5.333	5.737
0.25%	40%	all SNPs	0.910	3.707	2.882	5.333	1.136
1.0%	10%	max	0.695	0.800	0.773	0.766	0.374
1.0%	10%	all SNPs	0.069	0.498	0.102	0.740	0.056
1.0%	20%	max	1.439	1.513	1.350	1.206	1.253
1.0%	20%	all SNPs	0.042	0.915	0.132	1.153	0.040
1.0%	30%	max	3.018	3.189	3.274	3.369	2.656
1.0%	30%	all SNPs	0.274	3.189	1.059	3.367	0.730
1.0%	40%	max	2.906	3.104	3.011	3.407	3.404
1.0%	40%	all SNPs	0.599	2.147	1.565	3.407	1.178
2.5%	10%	max	0.730	0.934	0.982	1.007	0.960
2.5%	10%	all SNPs	0.035	0.387	0.112	0.977	0.046
2.5%	20%	max	1.912	2.106	2.096	2.045	2.132
2.5%	20%	all SNPs	0.352	1.452	0.797	2.045	0.338
2.5%	30%	max	1.269	1.398	1.440	1.482	1.712
2.5%	30%	all SNPs	0.353	0.972	0.648	1.482	0.324
2.5%	40%	max	2.691	2.861	2.787	2.925	3.651
2.5%	40%	all SNPs	0.706	2.043	1.554	2.924	0.605

A mixture of small and large effects was simulated as described in the main text. The best performing PRS weighting method in each scenario is in bold. % causal: percentage of causal markers; h²: total heritability explained by panel markers. For all methods except LDpred, we first applied LD-clumping with an r² threshold of 0.25 to all SNPs. “Max” refers to the maximum prediction R² achieved after optimizing over a range of p-value thresholds or fractions of causal variants. “All.SNPs” refers to the predictive performance using all SNPs after LD-clumping, except for LDpred where no clumping was performed. All predictive performances were measured by R² in %.

Table 3. Predictive performances (prediction R 2 in %) of the standard PRS and four other PRS schemes in simulations using real genotype data (a mixture small and large effects simulated).

Table 3. Predictive performances (prediction R ² in %) of the standard PRS and four other PRS schemes in simulations using real genotype data (a mixture small and large effects simulated).