Table 4 Generalization analysis of our method

From: A collaborative large language model for drug analysis

Method

Class

Acids

Lipids

Benzenoids

Oxygen

Phenylpropanoids

Nitrogen

Organoheterocyclic

Overall

GPT-4-06132

85.7

89.4

84.4

83.7

92.7

87.2

86.1

87.3

GPT-4-11062

90.5

87.2

86.7

88.4

92.7

94.9

97.2

91.4

GPT-4-01252

93.7

85.1

91.1

90.7

90.2

89.7

91.7

91.2

DrugGPT-G

88.9 (↓ 6.3)

89.4 (↓ 2.1)

84.4 (↓ 4.5)

86.0 (↓ 4.7)

92.7 (–)

87.2 (↓ 7.7)

86.1 (↓ 11.1)

88.5 (↓ 5.0)

DrugGPT

95.2

91.5

88.9

90.7

92.7

94.9

97.2

93.5

Method

Type

Mechanism of action

Small molecule

Biotech

Overall

Inhibitor

Antagonist

Overall

GPT-4-06132

87.7

85.7

87.3

86.8

89.5

87.3

GPT-4-11062

91.7

90.5

91.4

86.8

94.7

91.4

GPT-4-01252

90.6

93.7

91.2

81.6

94.7

91.2

DrugGPT-G

88.4 (↓ 4.7)

88.9 (↓ 6.3)

88.5 (↓ 5.0)

89.5 (↓ 2.6)

89.5 (↓ 5.2)

88.5 (↓ 5.0)

DrugGPT

93.1

95.2

93.5

92.1

94.7

93.5

  1. DrugGPT-G and DrugGPT indicate our method tested on data that is distinct from or from the same classes/types/MoAs as the training set, respectively. (↓ number) denotes the decrease performance in DrugGPT-G compared with DrugGPT. Bold values indicate the highest performance for each dataset.
Back to article page