Fig. 6From: An exploration strategy improves the diversity of de novo ligands using deep reinforcement learning: a case for the adenosine A2A receptorComparison of the properties of generated molecules by the pre-trained (a) and fine-tuned models (b) and molecules in the ZINC set (a) and the A2AR set (b), respectively. These properties included the number of hydrogen bond acceptors/donors, rotatable bonds, aliphatic rings, aromatic rings, and heterocyclesBack to article page