GNN-Retro: Retrosynthetic Planning with Graph Neural Networks

Retrosynthetic planning plays an important role in the field of organic chemistry, which could generate a synthetic route for target product. The is series reactions are started from available molecules. most challenging problem generation large search space candidate reactions. Estimating cost has been proved effectively to prune space, achieve higher accuracy with same iteration. And estimation one reaction comprised estimations all its reactants. So, how estimate these reactants will directly influence quality results. To get better performance, we propose new framework, named GNN-Retro, retrosynthetic by combining graph neural networks(GNN) and latest algorithm. structure GNN our framework incorporate information neighboring molecules, improve framework. experiments on USPTO dataset show that outperform state-of-the-art methods margin under settings.