Join Marwin Segler, Machine Learning Researcher at BenevolentAI for a talk on Automated Synthesis Planning with Machine Learning.
Computer-aided retrosynthesis, also known as automated synthesis planning (ASP) promises to be a valuable tool to find better synthetic routes faster and to determine the synthesizability of virtually designed compounds. However, despite of decades of research [1,2], ASP was never widely accepted by chemists, because the systems were slow, and the results were considered to be of unsatisfactory quality [3,4,5].
In this talk, we present our recent work on retrosynthesis using machine learning and modern search algorithms [6,7]. First, we show that deep neural networks can be trained over night on essentially all published chemical reactions to predict and rank the most suitable transforms to make a molecule . This allows the machine to learn the tolerated and conflicting functional groups and the scope of reactions . Second, to perform search, we employ Monte Carlo Tree Search (MCTS). MCTS allows to efficiently treat problems with very large branching factors, and does not rely strongly on hand-designed search heuristics, which makes it very well suited for retrosynthesis .
In comparison to the stateof the art search technique, Best First Search with hand- coded heuristics , our approach solves twice as many molecules and is almost two orders of magnitudes faster . To assess the quality of the predicted routes, we conducted double blind tests. Here, we found, for the first time, that organic chemists could not distinguish between real routes taken from the literature and predicted routes . We will close the talk by discussing implications, limitations and potential futures lines of research.
 G. Vleduts, Information Storage and Retrieval, 1963, 117
 E.J. Corey, W.T. Wipke. Science, 1969, 166, 178
 W.D. Ihlenfeldt, J. Gasteiger, Angew. Chem. Int. Ed., 1996, 34, 2613  S. Szymkuc et al., Angew. Chem. Int. Ed., 2016, 55, 5904
 A. Cook et al., W. Interd. Rev. Comp. Mol. Sci., 2012, 79
 M. Segler, M. P. Waller, Chem. Eur. J. 2017, 23, 5966
 M. Segler, M. Preuß, M. P. Waller, Nature 2018, 555, 604
Marwin Segler is Machine Learning Researcher at BenevolentAI. He studied chemistry at Westfälische Wilhelms-Universität Münster, Germany. Duringhis MSc thesis in synthetic organic chemistry in the group of Prof. Olga Garcia-Mancheño, he worked at the bench on C-H functionalization. For his PhD, still in Münster, Marwin joined the group of Prof. Mark Waller to study organic chemistry from a computational perspective. He investigated approaches for the computer-aided invention of novel chemical reaction types, introduced neural generative models to de-novo drug design, and introduced machine-learning to computer-aided retrosynthesis. Marwin now works at BenevolentAI in the Chemistry group.