Table 2 Module performance comparison on reaction center identification and synthon completion in top-k accuracy (%).

From: G2Retro as a two-step graph generative models for retrosynthesis prediction

Module

Method

Coverage (%)

Reaction type known

Reaction type unknown

   

1

2

3

5

1

2

3

5

Reaction center identification

G2G

97.9

90.2 (92.1)

94.5 (96.5)

94.9 (96.9)

95.0 (97.0)

75.8 (77.4)

83.9 (85.7)

85.3 (87.1)

85.6 (87.4)

 

GraphRetro

95.0

84.6 (89.1)

92.2 (97.1)

93.7 (98.6)

94.5 (99.5)

70.8 (74.5)

85.1 (89.6)

89.5 (94.2)

92.7 (97.6)

 

RetroPrime

100.0

84.6 (84.6)

94.0 (94.0)

96.7 (96.7)

97.9 (97.9)

65.6 (65.6)

81.3 (81.3)

87.7 (87.7)

92.0 (92.0)

 

G2Retro

97.5

84.3 (86.5)

94.6 (97.0)

96.5 (99.0)

97.0 (99.5)

69.5 (71.3)

85.6 (87.8)

90.8 (93.1)

94.8 (97.2)

 

G2Retro-B

97.5

85.0 (87.2)

94.1 (96.5)

96.2 (98.7)

97.3 (99.8)

69.3 (71.1)

85.4 (87.6)

91.1 (93.4)

94.7 (97.1)

Synthon completion

G2G

100.0

66.8

-

87.2

91.5

61.1

-

81.5

86.7

 

GraphRetro

99.7

77.4 (77.6)

89.5 (89.8)

94.2 (94.5)

97.6 (97.9)

75.6 (75.8)

87.4 (87.7)

92.5 (92.8)

96.1 (96.4)

 

RetroPrime

100.0

75.0

-

88.9

90.6

73.4

-

87.9

89.8

 

G2Retro

100.0

72.8

85.6

90.2

93.0

73.3

84.6

89.6

92.8

  1. Columns with 1, 3, 5 and 10 present top-1, top-3, top-5 and top-10 accuracies, respectively. Column “Coverage(%)” represents the percentage of test reactions that the modules of methods can be applied to. “(  )”: the accuracy within the covered reactions. All the baseline results are reported in their original papers, where “-” represents that the corresponding results are not reported.