Fig. 7From: An exploration strategy improves the diversity of de novo ligands using deep reinforcement learning: a case for the adenosine A2A receptorThe chemical space of generated molecules by the pre-trained model with the ZINC set (a–c) and the fine-tuned model with the A2AR set (d–f). The chemical space was represented by either logP ~ MW (a, d), first two components in PCA on PhysChem descriptors (c, e), and t-SNE on ECFP6 fingerprints (d, f)Back to article page