. Author manuscript; available in PMC: 2014 Jan 1.

Published in final edited form as: Methods Mol Biol. 2013;972:121–139. doi: 10.1007/978-1-60327-337-4_8

Table 2.

Comparison of performance in terms of sensitivity, specificity and false discovery rate (FDR) of three different procedures based on 100 replications for four different scenarios with different percentages of DE genes (q).

q	Method	Sensitivity	Specificity	FDR
	HMRF	0.80(0.029)	1.00(0.0023)	0.045(0.019)
0.115 (0.005)	HMRF-I	0.70(0.042)	0.99(0.0027)	0.079(0.025)
0.115 (0.005)	EB	0.69(0.054)	0.99(0.0027)	0.079(0.05)

	HMRF	0.87(0.033)	0.99(0.0049)	0.058(0.020)
0.189 (0.008)	HMRF-I	0.76(0.03)	0.99(0.004)	0.074(0.018)
0.189 (0.008)	EB	0.75(0.032)	0.99(0.0041)	0.075(0.018)

	HMRF	0.91(0.016)	0.97(0.0065)	0.054(0.010)
0.357 (0.009)	HMRF-I	0.84(0.020)	0.97(0.0063)	0.066(0.011)
0.357 (0.009)	EB	0.83(0.022)	0.97(0.0064)	0.066(0.011)

	HMRF	0.95(0.012)	0.94(0.012)	0.061(0.012)
0.486 (0.008)	HMRF-I	0.88(0.015)	0.95(0.0086)	0.060(0.0093)
0.486 (0.008)	EB	0.88(0.015)	0.95(0.0087)	0.060(0.0094)

HMRF: the proposed HMFR model using the network structures; HMRF-I: the proposed HMFR model without using the network structures; EB: the empirical Bayes method of Tai and Speed (2006) with FDRs matched to the HMRF algorithm; Summaries are averaged over 100 simulations; standard deviation is shown in parentheses.