Skip to main content

Table 8 Results of the experimental comparison between the proposed MODAS method and competing structural class prediction methods on the D1189 dataset.

From: Modular prediction of protein structural classes from sequences of twilight-zone identity with predicting sequences

Classifier used (name of the method, if any)

Feature vector

Reference

Accuracy

   

α

β

α/β

α+β

overall

SVM

AA composition, autocorrelations, and physicochemical properties

73

-

-

-

-

52.1

Bayesian classifier

AA composition

81

54.8

57.1

75.2

22.2

53.8

Logistic regression

AA composition, autocorrelations, and physicochemical properties

73

60.2

60.5

55.2

33.2

53.9

SVM

AA and polypeptide composition, physicochemical properties

45

-

-

-

-

54.7

Nearest neighbor

Pseudo-amino acid composition

67

48.9

59.5

81.7

26.6

56.9

Ensemble

AA composition, autocorrelations, and physicochemical properties

72

-

-

-

-

58.9

Nearest neighbor

Composition of tripeptides

52

-

-

-

-

59.9

IB1

PSI Blast based collocated AA pairs

75

65.3

67.7

79.9

40.7

64.7

Discriminant analysis

custom

78

62.3

67.7

63.1

66.5

65.2

SVM with RBF kernel (SCEC)

PSI Blast based collocated AA pairs

75

75.8

75.2

82.6

31.8

67.6

SVM with RBF kernel (SCPRED)

custom

79

89.1

86.7

89.6

53.8

80.6

SVM with polynomial or RBF kernels (MODAS)

custom

this paper

92.3

87.1

87.9

65.4

83.5

  1. The results were obtained using jackknife test. The methods are ordered by their average accuracies. Best results are shown in bold and "---" indicates results that were not reported by the original authors and which cannot be duplicated.