Decomposing Retrosynthesis into Reaction Center Prediction and Molecule Generation
Xianggen Liu, Pengyong Li, Hailin Hu, Jianyang Zeng, and Sen Song
Received Date: 1st August 19
Chemical retrosynthesis has been a crucial and challenging task in organic chemistry for several decades. In early years, retrosynthesis is mainly accomplished by the disconnection approach which is often labor-intensive and requires expert knowledge. Afterward, rule-based methods have dominated in retrosynthesis for years. In this study, we integrate the theory of disconnection approach into deep learning (DL) to boost the prediction performance and show that it can also increase the explainability of DL. Concretely, we propose a novel graph-based deep-learning framework, named DeRetro, to predict the set of reactants for a target product by executing the process of disconnection and reactant generation orderly. Experimental results report that DeRetro achieves new state-of-the-art performance in predicting the reactants. In-depth analyses also demonstrate that even without the reaction types as input, DeRetro maintains its retrosynthesis performance while other methods show a significant decrease, resulting in 19% accuracy margin between DeRetro and previous state-of-the-art rule-based methods. These results have established DeRetro as a powerful and useful computational tool in both reaction center prediction and retrosynthetic reaction prediction.
Read in full at bioRxiv.
This is an abstract of a preprint hosted on an independent third party site. It has not been peer reviewed but is currently under consideration at Nature Communications.