Research Article

Robust Significance Analysis of Microarrays by Minimum β-Divergence Method

Table 2

Performance evaluation of different methods using average values of FDR associated with varying proportions of DE gene.

Percentage of DE geneANOVAKWSAMLIMMAEBGAGABRIDGEProposed

Average FDR for small-sample case

2%0.3550.2050.2150.2100.2600.4050.3800.215
(0.755)(0.710)(0.735)(0.740)(0.785)(0.780)(0.485)(0.210)
[0.650][0.600][0.630][0.620][0.820][0.680][0.625][0.235]

4%0.2800.1520.1650.1680.2220.3470.3220.165
(0.748)(0.720)(0.720)(0.720)(0.785)(0.775)(0.390)(0.168)
[0.595][0.575][0.572][0.572][0.792][0.632][0.398][0.200]

6%0.2870.1970.2050.1980.2520.3670.3420.200
(0.738)(0.725)(0.725)(0.728)(0.798)(0.782)(0.372)(0.207)
[0.578][0.575][0.575][0.573][0.787][0.653][0.390][0.227]

8%0.2660.1810.1960.1910.2310.3180.2960.196
(0.728)(0.720)(0.710)(0.719)(0.669)(0.761)(0.329)(0.195)
[0.578][0.570][0.570][0.570][0.774][0.640][0.346][0.210]

10%0.2390.1520.1650.1560.1900.3030.2850.165
(0.585)(0.574)(0.563)(0.571)(0.569)(0.639)(0.305)(0.163)
[0.564][0.553][0.563][0.563][0.773][0.646][0.310][0.178]

Average FDR for large-sample case

2%0.0750.0800.0600.0600.1100.1150.1100.060
(0.500)(0.105)(0.470)(0.495)(0.570)(0.580)(0.145)(0.065)
[0.405][0.095][0.410][0.400][0.485][0.480][0.125][0.070]

4%0.0820.0870.0850.0850.1150.1170.1150.085
(0.510)(0.115)(0.458)(0.500)(0.540)(0.560)(0.148)(0.082)
[0.408][0.102][0.420][0.410][0.480][0.498][0.125][0.088]

6%0.0900.0920.0880.0860.1010.1110.110.088
(0.515)(0.118)(0.453)(0.510)(0.530)(0.533)(0.145)(0.088)
[0.400][0.113][0.403][0.398][0.460][0.467][0.125][0.087]

8%0.0700.0700.060.0720.0960.1150.1090.069
(0.460)(0.104)(0.384)(0.450)(0.459)(0.459)(0.134)(0.069)
[0.398][0.081][0.416][0.398][0.465][0.471][0.120][0.075]

10%0.0620.0630.0570.0570.0750.0910.0870.059
(0.361)(0.079)(0.299)(0.347)(0.355)(0.363)(0.104)(0.060)
[0.395][0.080][0.409][0.396][0.443][0.458][0.098][0.063]

In this table the values within the brackets (⋅), , and [⋅] represent the average values of FDR estimated by different methods (ANOVA, KW, SAM, LIMMA, EB, GAGA, BRIDGE, and proposed) in presence of one or two outliers in each of 10%, 20%, and 50% genes, respectively.