Fig. 4
From: Transformer-based molecular optimization beyond matched molecular pairs
Tanimoto similarity distribution for Similarity (≥ 0.5) dataset, Similarity ([0.5,0.7)) dataset, Similarity (≥ 0.7) dataset, MMP dataset, Scaffold dataset and Scaffold generic dataset. Legend Train for the molecular pairs from the training set; Generated desirable property for the pairs between the generated molecules that fulfil successful property constraints and their starting molecules from the test set; Generated desirable property+structure for the pairs between the generated molecules that fulfil both successful property and structure constraints and their starting molecules from the test set; Generated desirable property\(\lnot\)structure for the pairs between the generated molecules that fulfil successful property but not structure constraints and their starting molecules from the test set