Skip to main content

Table 6 The comparison between condtional generative models RL based models in the task of generating molecules satisfying condition \(C_1\) (that is, QED > 0.84 and SA score < 1.9)

From: Multi-objective de novo drug design with conditional graph generative model

Model

% valid

\(R_{\mathbf{c}}\)

\(EOR_{\mathbf{c}}\)

Diversity (\(I_{\mathbf{c}}\))

REINVENT

0.999 ± 0.001

0.986 ± 0.004

110

0.73 ± 0.07

Naive RL

0.993 ± 0.006

0.948 ± 0.052

105

0.64 ± 0.05

RL + Prior

1.000 ± 0.000

0.999 ± 0.000

111

0.44 ± 0.16

Graph

0.929 ± 0.003

0.73 ± 0.01

91

0.863 ± 0.001

SMILES

0.613 ± 0.015

0.66 ± 0.00

82

0.863 ± 0.000

  1. Results are reported as \(Mean \pm SD\)
  2. The best performance in each metric is highlighted in italics face