When designing experiments chemists face a complex multidimensional optimization problem and have to explore the nearly endless chemical space based on intuition acquired over years of training and experience. Although intensively studied, predicting the success of a chemical transformation remains a major challenge in Organic Chemistry and is often perceived as an art only human experts can do. Exploiting the analogies between human language and Organic Chemistry, neural machine translation models have recently outperformed templated-based methods and achieved state-of-the-art results in predicting the outcome of chemical reactions. In this work, we analyze the results of such fully data-driven neural sequence-to-sequence models, gain insight into the learned reaction mechanisms using a set of classified reactions and outline the potential of attention weight matrices for atom mappings generation.

“Found in Translation”: A deeper analysis of neural machine translation models for chemical reaction prediction

256th National Meeting (2018)

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