Table 1. Performance evaluation based on simulated gene expression profiles with m = 2 conditions/groups.

Results for the small-sample case (n1 = n2 = 3)
Methods	TPR	FPR	TNR	FNR	FDR	MER	AUC	pAUC	TPR	FPR	TNR	FNR	FDR	MER	AUC	pAUC
	Without outlying expressions								For 1 outlier with each of 5% genes
ANOVA	0.939	0.001	0.998	0.072	0.072	0.004	0.915	0.184	0.475	0.011	0.989	0.525	0.525	0.021	0.474	0.095
SAM	0.955	0.001	0.999	0.045	0.045	0.002	0.954	0.191	0.490	0.010	0.990	0.510	0.510	0.020	0.491	0.098
LIMMA	0.958	0.001	0.999	0.042	0.042	0.002	0.959	0.192	0.485	0.011	0.989	0.515	0.515	0.021	0.484	0.097
eLNN	0.932	0.001	0.999	0.068	0.068	0.003	0.931	0.186	0.372	0.013	0.987	0.627	0.627	0.025	0.371	0.074
EBarrays	0.938	0.001	0.999	0.062	0.062	0.002	0.939	0.188	0.307	0.014	0.986	0.692	0.692	0.028	0.306	0.061
BetaEB	0.938	0.001	0.999	0.062	0.062	0.002	0.937	0.188	0.940	0.001	0.999	0.060	0.060	0.002	0.941	0.188
KW	0.958	0.000	1.000	0.042	0.042	0.001	0.977	0.196	0.497	0.010	0.990	0.502	0.502	0.020	0.496	0.100
Proposed	0.939	0.001	0.998	0.072	0.072	0.004	0.915	0.184	0.936	0.001	0.998	0.064	0.064	0.004	0.914	0.182
	For 1 outlier with each of 10% genes								For 1 outlier with each of 75% genes
ANOVA	0.318	0.014	0.986	0.682	0.682	0.027	0.318	0.063	0.087	0.019	0.981	0.912	0.912	0.036	0.086	0.018
SAM	0.323	0.014	0.986	0.677	0.677	0.027	0.324	0.064	0.087	0.019	0.981	0.912	0.912	0.036	0.086	0.018
LIMMA	0.318	0.014	0.986	0.682	0.682	0.027	0.317	0.064	0.087	0.019	0.981	0.912	0.912	0.036	0.088	0.018
eLNN	0.250	0.015	0.985	0.750	0.750	0.030	0.251	0.050	0.025	0.020	0.980	0.975	0.975	0.039	0.026	0.005
EBarrays	0.210	0.016	0.984	0.790	0.790	0.032	0.211	0.042	0.025	0.020	0.980	0.975	0.975	0.039	0.024	0.005
BetaEB	0.940	0.001	0.999	0.060	0.060	0.002	0.941	0.188	0.025	0.020	0.980	0.975	0.975	0.039	0.024	0.005
KW	0.325	0.014	0.986	0.675	0.675	0.027	0.326	0.065	0.087	0.019	0.981	0.912	0.912	0.036	0.088	0.018
Proposed	0.932	0.001	0.998	0.098	0.098	0.004	0.907	0.181	0.920	0.001	0.998	0.080	0.080	0.004	0.887	0.177
Results for the large-sample case (n1 = n2 = 15)
Methods	TPR	FPR	TNR	FNR	FDR	MER	AUC	pAUC	TPR	FPR	TNR	FNR	FDR	MER	AUC	pAUC
Without outlying expressions								For 1 or 2 outliers with each of 5% genes
ANOVA	0.971	0.002	0.998	0.026	0.026	0.001	0.972	0.195	0.560	0.009	0.991	0.440	0.440	0.018	0.561	0.112
SAM	0.978	0.000	1.000	0.022	0.022	0.001	0.979	0.196	0.613	0.008	0.992	0.388	0.388	0.015	0.612	0.122
LIMMA	0.978	0.000	1.000	0.022	0.022	0.001	0.977	0.196	0.562	0.009	0.991	0.438	0.438	0.018	0.563	0.112
eLNN	0.975	0.001	0.999	0.025	0.025	0.001	0.974	0.195	0.850	0.003	0.997	0.150	0.150	0.006	0.849	0.170
EBarrays	0.973	0.001	0.999	0.028	0.028	0.001	0.974	0.195	0.562	0.009	0.991	0.438	0.438	0.018	0.561	0.112
BetaEB	0.973	0.001	0.999	0.028	0.028	0.001	0.975	0.195	0.973	0.001	0.999	0.028	0.028	0.001	0.972	0.195
KW	0.973	0.001	0.999	0.028	0.028	0.001	0.973	0.194	0.911	0.002	0.998	0.089	0.089	0.008	0.801	0.160
Proposed	0.971	0.002	0.998	0.026	0.026	0.001	0.973	0.195	0.975	0.001	0.999	0.025	0.025	0.001	0.976	0.195
For 1 or 2 outliers with each of 10% genes								For 1 or 2 outliers with each of 75% genes
ANOVA	0.420	0.012	0.988	0.580	0.580	0.023	0.420	0.084	0.312	0.014	0.986	0.688	0.688	0.028	0.311	0.062
SAM	0.497	0.010	0.990	0.502	0.502	0.020	0.497	0.099	0.347	0.013	0.987	0.652	0.652	0.026	0.347	0.069
LIMMA	0.425	0.012	0.988	0.575	0.575	0.023	0.425	0.085	0.347	0.013	0.987	0.652	0.652	0.026	0.348	0.069
eLNN	0.885	0.002	0.998	0.115	0.115	0.005	0.885	0.177	0.855	0.003	0.997	0.145	0.145	0.006	0.856	0.171
EBarrays	0.427	0.012	0.988	0.573	0.573	0.023	0.427	0.085	0.228	0.016	0.984	0.772	0.772	0.031	0.227	0.045
BetaEB	0.973	0.001	0.999	0.028	0.028	0.001	0.973	0.195	0.225	0.016	0.984	0.775	0.775	0.031	0.224	0.045
KW	0.851	0.004	0.996	0.149	0.149	0.008	0.807	0.161	0.808	0.009	0.991	0.192	0.192	0.002	0.936	0.187
Proposed	0.973	0.001	0.999	0.028	0.028	0.001	0.973	0.195	0.978	0.000	1.000	0.022	0.022	0.001	0.979	0.196

Average performance results of eight methods (ANOVA, SAM, LIMMA, eLNN, EBarrays, BetaEB, KW and Proposed) based on 100 datasets generated using a one-way ANOVA model with m = 2 groups/conditions and σ ² = 0.05 for both sample sizes n1 = n2 = 3 and n1 = n2 = 15. Each dataset for each case contained 300 true DE genes, and the remainder were 19700 true EE genes. The performance indices/measures TPR, FPR, TNR, FNR, FDR, MER and AUC were calculated for each method based on the top 300 estimated DE genes, under the assumption that the other estimated genes in each dataset for each case were EE genes for each method. The performance measure ‘pAUC’ was calculated at FPR = 0.2 for each method and for each dataset.