Predicting Retrosynthetic Reactions Using Self-Corrected Transformer Neural Networks

• DOI: 10.1021/acs.jcim.9b00949
• Publication Date: 2019-12-11T20:28:36Z
• Authors: Shuangjia Zheng; Jiahua Rao; Zhongyue Zhang; Jun Xu; Yuedong Yang
• Keywords: Retrosynthetic analysis; Benchmark (surveying); Training set

Synthesis planning is the process of recursively decomposing target molecules into available precursors. Computer-aided retrosynthesis can potentially assist chemists in designing synthetic routes; however, at present, it cumbersome and cannot provide satisfactory results. In this study, we have developed a template-free self-corrected predictor (SCROP) to predict using transformer neural networks. method, was converted machine translation problem from products molecular linear notations reactants. By coupling with network-based syntax corrector, our method achieved an accuracy 59.0% on standard benchmark data set, which outperformed other deep learning methods by >21% template-based >6%. More importantly, 1.7 times more accurate than state-of-the-art for compounds not appearing training set.