Planning Chemical Syntheses with Deep Neural Networks and Discipline Scale Data

Mark Waller

Since E. J. Corey first proposed to use a computer to assist in traversing deep synthetic trees in the late 1960s, chemists have tried to algorithmically discover the rules of chemistry. We have recently demonstrated how well neural networks combined with Monte Carlo Tree Search (MCTS) perform when planning complex retrosynthetic routes. We have validated our approach on 497 randomly selected targets, and performed double blind AB testing.

We will discuss the importance of developing fast and accurate retrosynthetic approaches as de novo molecular design becomes increasingly more important in modern drug discovery.

Back to Workshop II: Interpretable Learning in Physical Sciences