Skip to main content

Table 4 Performance comparison of Transformer and baselines in terms of successful property constraints, successful structure constraints and both metrics simultaneously

From: Transformer-based molecular optimization beyond matched molecular pairs

Dataset

Model

Successful property constraints (%)

Successful structure constraints (%)

Successful property and structure constraints (%)

MMP

Transformer

61.90

91.55

58.09

Transformer-U

33.67

93.25

31.85

Random

13.44±0.43

100

13.44±0.43

Similarity (\(\ge\)0.5)

Transformer

51.83

82.30

44.53

Transformer-U

29.04

83.63

25.32

Random

15.17±0.27

100

15.17±0.27

Similarity ([0.5,0.7))

Transformer

46.75

68.09

32.96

Transformer-U

26.23

69.13

18.72

Random

14.57±0.37

100

14.57±0.37

Similarity (\(\ge\)0.7)

Transformer

65.09

82.68

56.07

Transformer-U

39.57

84.83

34.70

Random

11.48±0.29

100

11.48±0.29

Scaffold

Transformer

61.53

95.32

59.69

Transformer-U

37.16

95.69

36.26

Random

17.22±0.74

100

17.22±0.74

Scaffold generic

Transformer

55.05

96.01

53.66

Transformer-U

32.55

96.30

31.69

Random

16.48±0.41

100

16.48±0.41

  1. The results in bold indicate the best values; higher values are better
  2. Each model is trained on the corresponding dataset for that row