Skip to main content

Table 6 Pathway and pathway pair classifiers

From: Iterative feature removal yields highly discriminative pathways

  

Single Pathways

Pathway Pairs

Sparse SVM

Data Set

IFR Iters

Val.( μ/σ)

Test

Top5

Val.( μ/σ)

Test

Top5

Val.( μ/σ)

Test

Top5

Lung

30

95.5/7.0

99.3

96.9

98.8/3.7

98.0

98.8

99.8/1.7

98.0

96.8

Influenza 11-14

30

98.6/2.9

96.0

95.5

99.0/2.3

97.3

97.6

98.9/2.3

92.0

92.8

Prostate

30

78.9/14.2

79.4

77.6

81.8/14.0

79.4

78.2

91.0/5.7

79.4

80.6

BCell Lymph.

20

89.5/12.3

83.3

75.0

91.8/12.4

83.3

78.3

87.8/12.4

83.3

80.0

  1. Cross-validated classification results compared using features selected from discriminative pathways and pathway-pairs, and the “optimal” set of features selected by sparse SVM (SSVM). The latter is equivalent to the first iteration of IFR. The mean and standard deviation of 50 cross-validated trials are shown, followed by the accuracy of the best model from the validation trials applied to withheld test data. Also shown is the average accuracy of the top 5 models on the withheld test data.