Skip to main content

Table 2 Comparison of the percentage of important substructures contained in the molecules generated by the different methods and the molecules in the ZINC and A2AR sets

From: An exploration strategy improves the diversity of de novo ligands using deep reinforcement learning: a case for the adenosine A2A receptor

  Fused ring (%) Furan ring (%) Benzene ring (%)
DrugEx (Pre-trained) 9.12 82.32 61.48
DrugEx (Fine-tuned) 60.69 66.35 65.62
REINVENT 0.20 95.26 61.98
ORGANIC 0.02 99.96 39.45
Pre-trained 24.22 4.51 63.31
Fine-tuned 76.33 23.82 72.85
ZINC 26.66 3.86 63.97
A2AR
 Active 79.09 40.29 75.33
 Inactive 76.73 9.33 70.88
  1. These methods contained DrugEx with pre-trained and fine-tuned model as different Gφ (in the parentheses), REINVENT, ORGANIC, Pre-trained model, and Fine-tuned model