Table 4. Performance evaluation based on simulated gene expression profiles with m ≥ 3 conditions with multiple patterns based on raw p-values and adjusted p-values (controlling FDR and FWER) at 1% for a large sample size.
| METHODS | Without outlying expressions | With 5% outlying expressions in 5% genes | With 5% outlying expressions in 10% genes |
|---|---|---|---|
| Using raw p-values | |||
| TPR.ANOVA | 0.981 | 0.631 | 0.503 |
| (0.010) | (0.009) | (0.009) | |
| TPR.SAM | 0.980 | 0.841 | 0.719 |
| (0.004) | (0.032) | (0.056) | |
| TPR.LIMMA | 0.983 | 0.635 | 0.508 |
| (0.010) | (0.009) | (0.009) | |
| TPR.KW | 0.982 | 0.975 | 0.975 |
| (0.008) | (0.008) | (0.008) | |
| TPR.Proposed | 0.983 | 0.981 | 0.981 |
| (0.037) | (0.036) | (0.036) | |
| Using adjusted p-values with Benjamini-Hochberg (BH) procedure (controlling FDR at 1%) | |||
| TPR.ANOVA.BH | 0.975 | 0.522 | 0.381 |
| (0.000) | (0.000) | (0.000) | |
| TPR.SAM.BH | 0.957 | 0.602 | 0.450 |
| (0.000) | (0.000) | (0.000) | |
| TPR.LIMMA.BH | 0.977 | 0.525 | 0.386 |
| (0.000) | (0.000) | (0.000) | |
| TPR.KW.BH | 0.972 | 0.872 | 0.855 |
| (0.000) | (0.000) | (0.000) | |
| TPR.Proposed.BH | 0.977 | 0.968 | 0.975 |
| (0.002) | (0.002) | (0.002) | |
| Using adjusted p-values with Bonferroni correction procedure (controlling FWER at 1%) | |||
| TPR.ANOVA.Bonf | 0.965 | 0.485 | 0.345 |
| (0.000) | (0.000) | (0.000) | |
| TPR.SAM.Bonf | 0.937 | 0.477 | 0.333 |
| (0.000) | (0.000) | (0.000) | |
| TPR.LIMMA.Bonf | 0.970 | 0.485 | 0.345 |
| (0.000) | (0.000) | (0.000) | |
| TPR.KW.Bonf | 0.941 | 0.581 | 0.469 |
| (0.000) | (0.000) | (0.000) | |
| TPR.Proposed.Bonf | 0.969 | 0.955 | 0.968 |
| (0.000) | (0.000) | (0.000) | |
Average performance results for five methods (ANOVA, SAM, LIMMA, KW and Proposed) based on 100 datasets generated using a one-way ANOVA model for the large-sample (n 1 = n2 = n 3 = n 4 = 15) case with m = 4 groups/conditions and σ 2 = 0.05, where each dataset contained 300 true genes and the remainder were 19700 true EE genes. The values represent the average TPRs based on raw p-values and adjusted p-values for the ANOVA, SAM, LIMMA, KW and Proposed methods in both the absence and the presence of outlying expressions. The value within the bracket () indicates the average FPRs. The adjusted p-values were calculated using Benjamini-Hochberg (BH) and Bonferroni correction methods.