Reaction condition recommendation is an essential element in synthetic planning, since reaction conditions are required for experimental validation, and properly designed reaction conditions can promote desired reactivity. However, it relies heavily on chemists’ knowledge and experience, and computer aided condition recommendation remains a challenging and under-explored problem. In this work, we develop a neural network model to predict the chemical context (solvent, reagent and catalyst) as well as temperature for any given organic reaction. A hierarchical model structure is designed to implicitly model compatibility, and teacher forcing is applied to improve training efficiency and accuracy. Trained on ~10 million examples from Reaxys, the model is able to predict a closely similar context combination, based on strictly defined criteria, to recorded catalyst, solvent and reagent in the top-ten predictions with an accuracy of 65%, with accuracies for individual elements reaching 80%~90%. Empirical evaluation shows that many of the failed predictions also show good functional similarity, or even are potential alternatives to the true conditions. Temperature is predicted to be within 20K deviation from the recorded temperature in 60%~70% of the test cases, with dependence on the correctness of chemical context prediction. The reaction condition information generated by this tool can be used to aid experimental design, improve the accuracy of in silico evaluation of reactivity, and pathway-level assessment and improvement of chemical synthesis processes.<br/>

Using machine learning to recommend suitable conditions for organic reactions

256th National Meeting (2018)

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