Development of reaction prediction system based on informatics and quantum chemical calculation

Abstract

In the field of chemoinformatics, computer-aided reaction prediction systems have been developed. However, the systems have not been widely used by experimental chemists because of its low accuracy for predicting chemical reactions. Recently, a reaction prediction scheme, which is based on the machine learning method and utilizes topological information of molecules as descriptors, has shown high effectiveness for organic reactions. Nevertheless, the application has been limited due to lack of steric and electronic information of molecules. The present study has developed a novel scheme, which uses the machine learning with descriptors obtained from quantum chemical calculation, in order to realize an accurate prediction for reactions including arbitrary chemical species such as organometallic and ionic reactions. The accuracy of present system has been clarified to be close to that of the previous system for basic polar and radical organic chemical reactions. In this presentation, we will show the details of the system and the dependencies on conditions of quantum chemical calculations.