An Efficient Approach to Screening Epigenome-Wide Data

Table 4

Simulation results for selecting important variables among 2,000 candidates including surrogate variables across 400 subjects.

Statistics	Bon	FDR	TT	Bon	FDR	TT
Random error normally distributed
# incorrect	2	1	1	24	10	9
Sensitivity	0.851	0.949	0.941	0.761	0.968	0.965
Specificity	1	1	1	1	0.997	0.997
# incorrect	64	29	21	210	141	93
Sensitivity	0.682	0.968	0.947	0.479	0.948	0.864
Specificity	1	0.988	0.994	0.999	0.925	0.976
Random error distributed (df = 1)
# incorrect	4	3	3	47	19	18
Sensitivity	0.612	0.767	0.786	0.531	0.878	0.883
Specificity	1	1	1	1	0.997	0.997
# incorrect	111	42	42	289	146	141
Sensitivity	0.445	0.88	0.837	0.281	0.841	0.717
Specificity	1	0.99	0.995	0.999	0.949	0.983

Bon: Bonferroni, FDR: false discovery rate, and TT: training and testing. Cutoff percentage of 20%.