Quantum Chemical Reaction Prediction Method Based on Machine Learning
Quantum Chemical Reaction Prediction Method Based on Machine Learning
复制标题
DOI：
10.1246/bcsj.20200017
复制
发表时间：
2020-05-01
影响因子：
4
通讯作者：
Nakai, Hiromi
中科院分区：
文献类型：
--
作者：
Fujinami, Mikito;Seino, Junji;Nakai, Hiromi
提出了一种基于机器学习的量子化学反应预测（QC-RP）方法。描述符包含反应物中的原子信息，例如通过量子化学计算获得的电荷、分子结构和原子/分子轨道。QC-RP方法包括两个过程，即学习和预测。学习过程使用有机化学教科书中的1625个极性和95个自由基反应构建筛选和排名分类器。在预测过程中，筛选分类器区分反应性和非反应性原子，排名分类器按排名顺序提供反应性原子对。数值评估证实了高精度的筛选和排名分类器的预测程序。此外，对分类器的分析揭示了预测的重要描述符。
A quantum chemical reaction prediction (QC-RP) method based on machine learning was developed to predict chemical products from given reactants. The descriptors contain atomic information in reactants such as charge, molecular structure, and atomic/molecular orbitals obtained by the quantum chemical calculations. The QC-RP method involves two procedures, namely, learning and prediction. The learning procedure constructs screening and ranking classifiers using 1625 polar and 95 radical reactions in a textbook of organic chemistry. In the prediction procedure, the screening classifier distinguishes reactive and unreactive atoms and the ranking one provides reactive atom pairs in ranking order. Numerical assessments confirmed the high accuracies both of the screening and ranking classifiers in the prediction procedures. Furthermore, an analysis on the classifiers unveiled important descriptors for the prediction.