Reaction selectivity and yield prediction are important for chemical synthesis. Previous computational methods either rely on time-consuming Quantum Mechanism-based model, which is difficult for experimental chemists to use, or black-box deep learning model, which is not accurate enough for specific reactions. Herein, we tackled the above problems with knowledge-guided model. Using convenient physically meaningful descriptors, we have developed an explainable machine learning platform that has achieved high accuracies in different reaction types and tasks, including regioselectivity, siteselectivity, enantioselectivity and yield predictions. Our model can be easily applied to different chemical reactions and extended to out-of-sample reactions well. Relevant experimental verification work is being carried out in cooperation.

Knowledge-guided machine learning platform for reaction selectivity (site-, regio- and enantioselectivity) and yield prediction

ACS Spring 2022

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.