机器学习设计单步逆向合成反应的研究进展

Research Progress in Single-step Retrosynthesis Design via Machine Learning

在逆向合成分析的过程中,对特定的目标分子设计单步逆向合成反应是探寻最优有机合成路线的关键环节。随着机器学习(Machine learning)研究的兴起,很多研究者开始尝试利用机器学习方法设计单步逆向合成反应。相关研究主要集中在两方面:(1)研究化合物分子输入方法;(2)基于特定的分子输入,研究各类单步逆向合成反应预测模型的构建方法。本文首先综述了分子输入的三种主流方法;然后分别分析了基于这三种分子输入方法构建的单步逆向合成反应预测模型的研究实例;之后,总结了当前机器学习方法设计单步逆向合成反应研究中存在的问题,并给出了解决问题的思路;最后,对机器学习设计单步逆向合成反应的前景作出展望。

In retrosynthetic analysis,designing single-step retrosynthetic reaction is an essential procedure in searching for optimal organic synthetic route.With the development of machine learning,researchers aim to design single-step retrosynthetic reaction by machine learning.The related study is mainly focused on two aspects,the molecule input methods and different modeling approaches based on specific molecule input.First,three popular types of molecule input methods in single-step retrosynthesis design models are introduced.Then,based on three popular types of molecule input methods,single-step retrosynthesis models with related study cases are analyzed respectively.In addition,current problems are summarized on the basis of recent study.Meanwhile,corresponding ideas are proposed.At last,some prospects are made for further study in single-step retrosynthesis design via machine learning.