Skip to main content

Table 4 Impact of molecular similarity on prediction performance

From: Cytochrome P450 site of metabolism prediction from 2D topological fingerprints using GPU accelerated probabilistic classifiers

Data

Classifier

Bond

 

Top-3%

 

Top-2%

set

 

depth

 

TS1

TS2

 

TS1

TS2

2C9

PRW

5

 

85

85

 

82

85

6

 

83

85

 

81

85

RASCAL

5

 

85

85

 

78

85

6

 

81

85

 

77

85

2D6

PRW

5

 

90

92

 

85

75

6

 

91

94

 

86

79

RASCAL

5

 

90

90

 

84

75

6

 

87

88

 

84

83

3A4

PRW

5

 

83

82

 

80

78

6

 

84

81

 

80

79

RASCAL

5

 

80

76

 

76

67

6

 

79

75

 

72

63

  1. Table shows the classification performance in terms of top-2% and top-3% for test set 1 (TS1: 20% of each isoform data set selected at random) and test set 2 (TS2: the 50% of molecules in TS1 most dissimilar to the training data).