Skip to main content

Table 2 Comparison of TMpro methods with that of TMHMM on high resolution data set of 36 proteins from benchmark analysis.

From: Transmembrane helix prediction using amino acid property features and latent semantic analysis

 

Method ↓

Qok

Segment F Score

Segment Recall

Segment Precision

Q2

# of TM proteins misclassified as soluble proteins

High resolution proteins (36 TM proteins)

1

TMHMM

71

90

90

90

80

3

2

TMpro LC

61

94

94

94

76

0

3

TMpro HMM

66

95

97

92

77

0

4

TMpro NN

83

96

95

96

75

0

Without SVD

5

TMpro NN without SVD

69

94

95

93

73

0

  1. It can be seen that TMpro achieves high segment accuracy (F-score) even with a simple linear classifier. For a description of evaluation metrics see caption of Table 3. To demonstrate the requirement of singular value decomposition of features, the results obtained by directly using property count features without employing SVD are shown in the last row. It may be seen that the results are slightly poor compared to standard TMpro.