G-003
モンテカルロ木探索による特定の有機低分子化合物を出発点とした誘導体の生成手法の開発

◎恵利川大樹・安尾信明・関嶋政和（東工大）

Development of a new drug requires an enormous amount of time and cost. In drug discovery, molecular optimization that finds molecule with better properties is important.ChemTS is an existing molecular generative model based on MCTS (Monte Carlo tree search). ChemTS has succeeded in generating better molecules more efficiently than existing models. However, ChemTS could not specify starting molecule, which means it could not generate derivatives. In this paper, we present a generative model using MCTS and RNN (recurrent neural network) starting from specific molecule.As a result, our model generated molecules, whose QED is higher than 0.93, starting from molecules whose average QED is 0.63.