Table 2.

Number of predictive genes passing the given R ² threshold in the Geuvadis data and GenoExp data

threshold	Geuvadis data				GenoExp data
	Lasso	ENET	LMM	BSLMM	Lasso	ENET	LMM	BSLMM
0.05	2252	2262	2447	2567	1785	1414	1560	1758
0.10	1144	1145	1145	1266	831	788	734	826
0.20	420	422	383	466	315	309	276	323
0.30	161	162	152	178	156	148	124	160
0.40	75	75	65	76	70	70	56	70
0.50	33	33	25	32	36	32	27	37
0.60	14	14	12	14	25	21	20	24

There are 15,810 and 15,427 genes in the Geuvadis data and GenoExp data, respectively. It can be seen that in both data sets when the given R ² threshold is large (e.g. ≥0.30) the number of predictive genes passing that value in LMM is less than that of LASSO, ENET or BSLMM, implying that these highly predictive genes may have a sparse genetic architecture