Fig. 10From: An exploration strategy improves the diversity of de novo ligands using deep reinforcement learning: a case for the adenosine A2A receptorComparison of the chemical space of active ligands in the A2AR set and generated molecules by DrugEx (fine-tuned, a–c), DrugEx (pre-trained, d–f), REINVENT (g–i), and ORGANIC (j–l). Chemical Space was represented by logP ~ MW (a, d, g, j), the first two components in PCA on PhysChem descriptors (b, e, h, k), and t-SNE on ECFP6 fingerprints (c, f, i, l)Back to article page