An Efficient Approach to Screening Epigenome-Wide Data

Table 5

Simulation results for selecting important variables among 2,000 candidates including surrogate variables across 800 subjects.

Statistics	Bon	FDR	TT	Bon	FDR	TT
Random error normally distributed
# incorrect	2	44	5	26	309	36
Sensitivity	0.991	1	0.998	0.949	1	0.988
Specificity	1	0.976	0.998	0.996	0.807	0.981
# incorrect	64	29	32	210	141	130
Sensitivity	0.682	0.968	0.847	0.479	0.948	0.696
Specificity	1	0.988	0.999	0.999	0.925	0.995
Random error distributed (df = 1)
# incorrect	0	1	1	4	9	2
Sensitivity	0.975	0.994	0.993	0.961	0.998	0.99
Specificity	1	1	1	1	0.995	0.999
# incorrect	15	34	7	82	214	50
Sensitivity	0.928	0.998	0.98	0.803	0.994	0.924
Specificity	1	0.981	0.998	0.998	0.868	0.988

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