Skip to main content

Table 3 Results of constrained optimization for 800 validation molecules

From: MERMAID: an open source automated hit-to-lead method based on deep reinforcement learning

\(\delta\)

GCPN

Mol-CycleGAN

MERMAID

Improvement

Similarity

Success (%)

Improvement

Similarity

Success (%)

Improvement

Similarity

Success (%)

0.2

4.12 ± 1.19

0.34 ± 0.11

100

5.79 ± 2.35

0.30 ± 0.11

93.8

9.94 ± 2.74

0.23 ± 0.04

100.0

0.4

2.49 ± 1.30

0.47 ± 0.08

100

2.89 ± 2.08

0.52 ± 0.10

58.8

6.04 ± 2.29

0.42 ± 0.02

100.0

0.6

0.79 ± 0.63

0.68 ± 0.08

100

1.22 ± 1.48

0.69 ± 0.07

19.3

1.99 ± 1.74

0.62 ± 0.02

85.3

  1. The results of GCPN and Mol-CycleGAN are cited from Maziarka et al. [36]. The mean and standard deviation of improvement, similarity, and success rate of generated molecules are shown