Skip to main content

Table 5 Comparison of performance with simulated datasets with 100 true positives

From: Exploratory differential gene expression analysis in microarray experiments with no or limited replication

Cut-off points

PPV

NPV

Specificity

Sensitivity

Likelihood ratio

Sapir and Churchill [20]

0.5

0.54

1.00

0.99

0.85

70.46

0.6

0.56

1.00

0.99

0.80

75.78

0.7

0.56

1.00

0.99

0.78

76.31

0.8

0.56

1.00

0.99

0.73

76.43

0.9

0.55

0.99

0.99

0.64

73.45

0.95

0.55

0.99

0.99

0.60

71.62

0.99

0.54

0.99

0.99

0.53

70.29

0.998

0.57

0.99

0.99

0.48

79.57

0.9998

0.58

0.99

0.99

0.43

82.78

0.99998

0.57

0.99

1.00

0.34

78.04

Newton et al. [19]

0.43 (0.3)

0.62

1.00

0.99

0.76

96.50

0.67 (0.4)

0.61

0.99

0.99

0.70

92.84

1 (0.5)

0.61

0.99

0.99

0.66

91.60

2.33 (0.7)

0.61

0.99

0.99

0.59

95.17

4 (0.8)

0.60

0.99

0.99

0.52

91.28

5.67 (0.85)

0.60

0.99

0.99

0.50

90.42

9 (0.9)

0.61

0.99

0.99

0.49

94.33

19 (0.95)

0.62

0.99

0.99

0.49

97.48

99 (0.99)

0.63

0.99

1.00

0.44

101.00

499 (0.998)

0.60

0.99

1.00

0.36

89.52

Loguinov et al. (this work)

0.5

0.40

1.00

0.98

0.99

39.65

0.25

0.60

1.00

0.99

0.93

90.99

0.2

0.63

1.00

0.99

0.91

102.47

0.15

0.68

1.00

0.99

0.90

127.89

0.1

0.72

1.00

0.99

0.87

157.34

0.05

0.76

1.00

1.00

0.82

188.22

0.01

0.79

1.00

1.00

0.76

226.78

0.002

0.80

0.99

1.00

0.60

238.72

0.0002

0.88

0.99

1.00

0.44

437.65

0.00002

0.89

0.99

1.00

0.40

477.44

Chen et al. [17]

0.05

0.3

1.00

.97

1.00

37.30

0.01

0.45

1.00

.98

0.88

49.55

  1. The first column lists the cut-off points used for each method for the performance comparison on simulated data as described in the text. For Sapir and Churchill [20] the cut-offs correspond to posterior probabilities of being differentially expressed. Similarly for Newton et al. [19] they correspond to posterior odds (probabilities) of true differential expression. For this work, the cut-offs for q-values are shown. Chen et al. [17] have two cut-offs which are approximated with a polynomial fit and are not shown in Figure 24 because there are no approximations available for other cut-offs. PPV is positive predictive value which equals TP (true positive)/TP + FP (false positive). NPV is negative predictive value which equals TN (true negative)/TN + FN (false negative). Sensitivity is TP/(TP + FN). Specificity is TN/(FP + TN). The likelihood ratio (Bayes' factor) is Sensitivity/(1-Specificity). For computations we used simulated data shown in Figure 22.