Fig. 5
From: DrugEx v3: scaffold-constrained drug design with graph transformer-based reinforcement learning
The distribution of the QED score (A, C) and SA score (B, D) of desired ligands. Shown are the molecules generated from the ChEMBL set and LIGAND set and of molecules generated by four different generators. For the QED score, four generators had the same performance as the molecules in both ChEMBL set (A) and the LIGAND set (C). For the SA score, Graph Transformer did not outperform three other SMILES-based generators in the ChEMBL set (B) and even worse in the LIGAND set (D).