Planning chemical syntheses with deep neural networks and symbolic AI

By Marwin H. S. Segler, Mike Preuss, Mark P. Waller. 2017

1,608 citations recorded by the source corpus.

To plan the syntheses of small organic molecules, chemists use retrosynthesis, a problem-solving technique in which target molecules are recursively transformed into increasingly simpler precursors. Computer-aided retrosynthesis would be a valuable tool but at present it is slow and provides results of unsatisfactory quality. Here we use Monte Carlo tree search and symbolic artificial intelligence (AI) to discover retrosynthetic routes. We combined Monte Carlo tree search with an expansion policy network that guides the search, and a filter network to pre-select the most promising retrosynthetic steps. These deep neural networks were trained on essentially all reactions ever published in organic chemistry. Our system solves for almost twice as many molecules, thirty times faster than the traditional computer-aided search method, which is based on extracted rules and hand-designed heuristics. In a double-blind AB test, chemists on average considered our computer-generated routes to be equivalent to reported literature routes.

Subfields: Artificial Intelligence, Computational Theory and Mathematics, Materials Chemistry, Computational Chemistry & Drug Discovery

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computational Theory and Mathematics

In Materials Chemistry

In Computational Chemistry & Drug Discovery

Citations in the corpus

Learning to predict the cosmological structure formation (2018, background): “This new paradigm is also significantly impacting a variety of domains in the sciences, from biology (27, 28) to chemistry (29, 30) and physics (31, 32).”
Machine Learning Methods in Drug Discovery (2020, methodology): “The MCTS technique is ideal for this purpose because of the simulation’s ability to perform random continuous step searches without branching until optimal conditions and solutions are met [82,83].”
Enabling late-stage drug diversification by high-throughput experimentation with geometric deep learning (2023, methodology): “Among the various machine learning methods developed for chemical reaction planning, [23, 25, 26], graph neural networks (GNNs) have been successfully employed for retrosynthesis planning, regioselectivityand reaction product-prediction.”
Modern Hopfield Networks for Few- and Zero-Shot Reaction Prediction (2021, methodology): “In the seminal work of Segler et al. (2018), an NN model is trained to identify the correct reaction template for a given molecule, where 301,671 reaction templates have been extracted from a large database of reactions.”
State-of-the-art augmented NLP transformer models for direct and single-step retrosynthesis (2020, methodology): “[5], and the forward transformer score by Schwaller et al.”
SMILES Transformer: Pre-trained Molecular Fingerprint for Low Data Drug Discovery (2019, background): “…al. 2015; Xu et al. 2017), molecule generation and lead optimization (G´omez-Bombarelli et al. 2018; Jin, Barzilay, and Jaakkola 2018), retrosynthesis planning (Segler, Preuss, and Waller 2018; Schwaller et al. 2018), and compound-protein afﬁnity prediction ( ¨Ozt¨urk, ¨Ozg¨ur, and Ozkirimli 2018).”
Artificial Intelligence (AI) Applications in Drug Discovery and Drug Delivery: Revolutionizing Personalized Medicine (2024, background): “Therefore, AI-driven retrosynthesis planning is essential for accelerating route discovery in different contexts and automating the evaluation process [127,129].”
Artificial Intelligence For Natural Product Drug Discovery (2023, methodology): “…output of a machine learning algorithm towards user-defined regions of optimality via a predefined (computational) reward function, has shown promise in de novo design towards attractive regions of chemical space 153–155 , for rule-based organic chemistry and for retrosynthesis prediction 156–159 .”