An Efficient Approach to Screening Epigenome-Wide Data

Table 1

Simulation results for selecting important variables among 2,000 candidates with surrogate variables included.

Statistics	Bon	FDR	TT	Bon	FDR	TT
Random error normally distributed
# incorrect	0	1	1	3	8	2
Sensitivity	0.981	0.991	0.994	0.968	0.999	0.991
Specificity	1	1	1	1	0.996	1
# incorrect	13	32	7	78	213	46
Sensitivity	0.936	0.998	0.982	0.814	0.996	0.931
Specificity	1	0.982	0.998	0.998	0.868	0.989
Random error distributed (df = 1)
# incorrect	1	1	1	16	9	7
Sensitivity	0.892	0.958	0.963	0.837	0.982	0.941
Specificity	1	1	1	1	0.996	1
# incorrect	46	29	21	169	158	95
Sensitivity	0.771	0.981	0.906	0.582	0.969	0.790
Specificity	1	0.986	0.999	0.999	0.909	0.993

Bon: Bonferroni, FDR: false discovery rate, and TT: training and testing.