Table 2.

Test set prediction accuracy comparison between snpnet-2.0, snpnet, bigstatsr, BOLT-LMM and LDpred2

	snpnet-2.0	snpnet	bigstatsr	BOLT-LMM	LDpred2	Covariates only
High cholesterol (B)	0.72533	0.72531	0.72705	0.70236	0.71240	0.69261
Asthma (B)	0.61609	0.61608	0.62257	0.62638	0.61112	0.53540
Standing height (Q)	0.71096	0.71100	0.71632	0.72169	0.67515	0.53789
BMI (Q)	0.11408	0.11412	0.12223	0.12869	0.094198	0.010859
Other hypothyroidism (S)	0.75194	0.75205	0.73818	0.71908	0.72158	0.66073
Thyrotoxicosis (S)	0.71020	0.71021	0.70432	0.67106	0.69009	0.64888

Note: For binary response, the test metric is the area under the ROC curve (AUC). For quantitative response, the metric is the R-squared. For survival response, the metric is the C-index. The results of bigstatsr, $\mathtt{BOLT-LMM}$ and LDpred2 on survival responses are based on regularized logistic regression on the disease indicator with age as an additional covariate. The results of BOLT-LMM on binary data, and LDpred2 on both binary and quantitative data, are obtained by refitting a logistic or linear regression using the polygenic score and the covariates on the training set.