An Efficient Approach to Screening Epigenome-Wide Data

Table 3

Simulation results for selecting important variables among 2,000 candidates including surrogate variables across 200 subjects.

Statistics	Bon	FDR	TT	Bon	FDR	TT
Random error normally distributed
# incorrect	7	6	6	77	43	41
Sensitivity	0.292	0.423	0.475	0.227	0.605	0.641
Specificity	1	1	1	1	0.998	0.997
# incorrect	165	87	93	361	214	244
Sensitivity	0.175	0.616	0.571	0.099	0.565	0.432
Specificity	1	0.994	0.996	1	0.975	0.989
Random error distributed (df = 1)
# incorrect	9	8	7	88	62	55
Sensitivity	0.126	0.193	0.307	0.124	0.408	0.498
Specificity	1	1	1	1	0.998	0.997
# incorrect	182	123	119	380	274	288
Sensitivity	0.093	0.422	0.435	0.051	0.377	0.314
Specificity	1	0.996	0.997	1	0.985	0.992

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