Fig. 3
From: MERMAID: an open source automated hit-to-lead method based on deep reinforcement learning
Details of SMILES generation by MCTS (a) With the path to the non-terminal node as input, the RNN model generates a complete partial SMILES. Then, by inserting the generated partial SMILES into the seed SMILES, a new SMILES is generated. b Selecting seed SMILES that is replaced with the generated one.Nodes under the root node have the information of the part deleted from seed SMILES.The generation of partial SMILES begins from the grandchild nodes of the root node. c The optimization cycle is repeated multiple times, replacing one of the generated SMILES with a seed SMILES