Since E. J. Corey first proposed to use a computer for retrosyntehsis in the 1960s, chemists have tried to algorithmically discover the rules of chemistry. We recently demonstrated how well modern methods could perform when planning complex retrosynthetic routes. Our approach was built on top of two key technologies: firstly, we showed that reactions could be quickly predicted using a neural network trained on massive datasets of reactions.<sup> </sup>Secondly, the other key component was an algorithm known as Monte Carlo Tree Search (MCTS) which we adopted for deep retrosynthesis. We validated our approach on 497 randomly selected targets, and performed double blind AB testing.<br/><br/>We have also developed a novel method for reaction prediction and retrosynthesis based on link prediction. This means we are not limited to rules based approaches, so we can effectively discover brand new reactions that have no precedence in the literature.

Learning to plan chemical syntheses

256th National Meeting (2018)

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