Table 1.

Comparison of seven different methods in predicting gene expression levels in the GEUVADIS data

Threshold	ENET	BayesR	BVSR	LMM	MultiBLUP	rjMCMC	DPR
							VB	MCMC
0.10	1061	809	486	1195	1098	1013	1163	1280
0.20	449	338	142	403	299	321	389	467
0.30	182	170	48	162	110	123	155	194
0.40	78	84	24	76	46	47	70	86
0.50	37	35	10	33	16	19	32	38
0.60	15	14	4	14	5	9	12	17
0.70	2	3	1	3	1	2	2	3

To compare prediction performance, we counted the number of genes whose median R ² across 20 replicates in the test set is above a given R ² threshold. A larger number thus indicates better performance. For each given threshold, we colored the best method with red and the second best method with blue