Learning to Plan Chemical Syntheses

Abstract

From medicines to materials, small organic molecules are indispensable for human well-being. To plan their syntheses, chemists employ a problem solving technique called retrosynthesis. In retrosynthesis, target molecules are recursively transformed into increasingly simpler precursor compounds until a set of readily available starting materials is obtained. Computer-aided retrosynthesis would be a highly valuable tool, however, past approaches were slow and provided results of unsatisfactory quality. Here, we employ Monte Carlo Tree Search (MCTS) to efficiently discover retrosynthetic routes. MCTS was combined with an expansion policy network that guides the search, and an "in-scope" filter network to pre-select the most promising retrosynthetic steps. These deep neural networks were trained on 12 million reactions, which represents essentially all reactions ever published in organic chemistry. Our system solves almost twice as many molecules and is 30 times faster in comparison to the traditional search method based on extracted rules and hand-coded heuristics. Finally after a 60 year history of computer-aided synthesis planning, chemists can no longer distinguish between routes generated by a computer system and real routes taken from the scientific literature. We anticipate that our method will accelerate drug and materials discovery by assisting chemists to plan better syntheses faster, and by enabling fully automated robot synthesis.

References in corpus (6)

• Training Very Deep Networks
• Neural networks for the prediction organic chemistry reactions
• Modelling Chemical Reasoning to Predict Reactions
• End-to-End Differentiable Proving
• Move Evaluation in Go Using Deep Convolutional Neural Networks
• Generating Focussed Molecule Libraries for Drug Discovery with Recurrent Neural Networks

Cited by in corpus (118)

• On the Opportunities and Risks of Foundation Models
• Mastering Atari, Go, Chess and Shogi by Planning with a Learned Model
• A Generative Model for Inverse Design of Metamaterials
• Molecular Transformer - A Model for Uncertainty-Calibrated Chemical Reaction Prediction
• Combining Machine Learning and Computational Chemistry for Predictive Insights Into Chemical Systems
• GuacaMol: Benchmarking Models for De Novo Molecular Design
• Deep learning for molecular design - a review of the state of the art
• Physics-informed neural networks for solving Reynolds-averaged Navier$\unicode{x2013}$Stokes equations
• MolGAN: An implicit generative model for small molecular graphs
• Graphene nanoribbons for quantum electronics
• State-of-the-Art Augmented NLP Transformer models for direct and single-step retrosynthesis
• Mapping the Landscape of Artificial Intelligence Applications against COVID-19
• Autonomous discovery in the chemical sciences part I: Progress
• SELFIES and the future of molecular string representations
• Mol-CycleGAN - a generative model for molecular optimization
• Polymer Informatics: Current Status and Critical Next Steps
• Learning to Predict the Cosmological Structure Formation
• The Exploration of Chemical Reaction Networks
• Generative chemistry: drug discovery with deep learning generative models
• SMILES Transformer: Pre-trained Molecular Fingerprint for Low Data Drug Discovery
• Neural Network Potentials for Chemistry: Concepts, Applications and Prospects
• MechRetro is a chemical-mechanism-driven graph learning framework for interpretable retrosynthesis prediction and pathway planning
• Deep Learning for Feynman's Path Integral in Strong-Field Time-Dependent Dynamics
• Optical coherent dot-product chip for sophisticated deep learning regression
• A Universal Density Matrix Functional from Molecular Orbital-Based Machine Learning: Transferability across Organic Molecules
• A Deep Generative Model for Molecule Optimization via One Fragment Modification
• Towards the design of chemical reactions: Machine learning barriers of competing mechanisms in reactant space
• Tutorial: How to Train a Neural Network Potential
• Deeply learning molecular structure-property relationships using attention- and gate-augmented graph convolutional network
• Generative Chemical Transformer: Neural Machine Learning of Molecular Geometric Structures from Chemical Language via Attention
• Augmenting adaptive immunity: progress and challenges in the quantitative engineering and analysis of adaptive immune receptor repertoires
• Regression-clustering for Improved Accuracy and Training Cost with Molecular-Orbital-Based Machine Learning
• Three-Dimensionally Embedded Graph Convolutional Network (3DGCN) for Molecule Interpretation
• Alchemy: A Quantum Chemistry Dataset for Benchmarking AI Models
• AI-accelerated Discovery of Altermagnetic Materials
• Precursor recommendation for inorganic synthesis by machine learning materials similarity from scientific literature
• From Chess and Atari to StarCraft and Beyond: How Game AI is Driving the World of AI
• Learning To Navigate The Synthetically Accessible Chemical Space Using Reinforcement Learning
• ${\mathsf{G}}^2\mathsf{Retro}$ as a Two-Step Graph Generative Models for Retrosynthesis Prediction
• A Bayesian algorithm for retrosynthesis
• Autonomous decision making for solid-state synthesis of inorganic materials
• Reinforcement Learning Applications
• Probability-Density-Based Deep Learning Paradigm for the Fuzzy Design of Functional Metastructures
• Monte Carlo Tree Search: A Review of Recent Modifications and Applications
• Machine Learning - Driven Materials Discovery: Unlocking Next-Generation Functional Materials - A review
• Re-evaluating Retrosynthesis Algorithms with Syntheseus
• Latent Molecular Optimization for Targeted Therapeutic Design
• Deep reinforcement learning-designed radiofrequency waveform in MRI
• RetroComposer: Composing Templates for Template-Based Retrosynthesis Prediction
• Chemical-Reaction-Aware Molecule Representation Learning
• Models Matter: The Impact of Single-Step Retrosynthesis on Synthesis Planning
• Deep learning-based design of broadband GHz complex and random metasurfaces
• Screening and understanding Li adsorption on 2-dimensional metallic materials by learning physics
• Tackling the challenge of a huge materials science search space with quantum-inspired annealing
• Learning with Delayed Rewards -- A case study on inverse defect design in 2D materials
• Accurate Molecular-Orbital-Based Machine Learning Energies via Unsupervised Clustering of Chemical Space
• On the Predictive Power of Chemical Concepts
• RetroGraph: Retrosynthetic Planning with Graph Search
• Automatically Extracting Action Graphs from Materials Science Synthesis Procedures
• A Model to Search for Synthesizable Molecules
• Amortized Tree Generation for Bottom-up Synthesis Planning and Synthesizable Molecular Design
• Spiers Memorial Lecture: How to do impactful research in artificial intelligence for chemistry and materials science
• Application and Assessment of Deep Learning for the Generation of Potential NMDA Receptor Antagonists
• Beyond Games: A Systematic Review of Neural Monte Carlo Tree Search Applications
• Combining Q-Learning and Search with Amortized Value Estimates
• Extracting many-particle entanglement entropy from observables using supervised machine learning
• RetroGNN: Approximating Retrosynthesis by Graph Neural Networks for De Novo Drug Design
• DeepAtom: A Framework for Protein-Ligand Binding Affinity Prediction
• Neural-Guided Symbolic Regression with Asymptotic Constraints
• Automated Retrosynthesis Planning of Macromolecules Using Large Language Models and Knowledge Graphs
• Graph Neural Networks for the Prediction of Substrate-Specific Organic Reaction Conditions
• DirectMultiStep: Direct Route Generation for Multistep Retrosynthesis
• Decoding Drug Discovery: Exploring A-to-Z In silico Methods for Beginners
• Does Hessian Data Improve the Performance of Machine Learning Potentials?
• Barking up the right tree: an approach to search over molecule synthesis DAGs
• Discovery of structure-property relations for molecules via hypothesis-driven active learning over the chemical space
• Retro-BLEU: Quantifying Chemical Plausibility of Retrosynthesis Routes through Reaction Template Sequence Analysis
• Practical Massively Parallel Monte-Carlo Tree Search Applied to Molecular Design
• Compressing physical properties of atomic species for improving predictive chemistry
• AndroidEnv: A Reinforcement Learning Platform for Android
• Finding Optimal Pathways in Chemical Reaction Networks Using Ising Machines
• ImJoy: an open-source computational platform for the deep learning era
• Machine learning and behavioral economics for personalized choice architecture
• Compounding meta-atoms into meta-molecules with hybrid artificial intelligence techniques
• Retrosynthesis with Attention-Based NMT Model and Chemical Analysis of the "Wrong" Predictions
• MCTS Based Agents for Multistage Single-Player Card Game
• Transfer Learning for Scientific Data Chain Extraction in Small Chemical Corpus with BERT-CRF Model
• Evolutionary Retrosynthetic Route Planning
• Ab initio machine learning in chemical compound space
• A hybrid deep-learning approach for complex biochemical named entity recognition
• A comprehensive study on the prediction reliability of graph neural networks for virtual screening
• Modern Hopfield Networks for Few- and Zero-Shot Reaction Template Prediction
• An In-depth Summary of Recent Artificial Intelligence Applications in Drug Design
• Adaptive Warm-Start MCTS in AlphaZero-like Deep Reinforcement Learning
• Dataset of Solution-based Inorganic Materials Synthesis Recipes Extracted from the Scientific Literature
• Gradient Origin Networks
• World Programs for Model-Based Learning and Planning in Compositional State and Action Spaces
• Learning Space Partitions for Path Planning
• SMILES-X: autonomous molecular compounds characterization for small datasets without descriptors
• Uncertainty quantification of molecular property prediction using Bayesian neural network models
• SynRXN: An Open Benchmark and Curated Dataset for Computational Reaction Modeling
• BioNavi-NP: Biosynthesis Navigator for Natural Products
• Self-Improved Retrosynthetic Planning
• Auto-Encoding Molecular Conformations
• High-Freedom Inverse Design with Deep Neural Network for Metasurface Filter in the Visible
• Efficient interpolation of molecular properties across chemical compound space with low-dimensional descriptors
• FragmentRetro: A Quadratic Retrosynthetic Method Based on Fragmentation Algorithms
• Accelerating Copolymer Inverse Design using AI Gaming algorithm
• To Root Artificial Intelligence Deeply in Basic Science for a New Generation of AI
• ChemiRise: a data-driven retrosynthesis engine
• Rapid Task-Solving in Novel Environments
• Predictive Synthesis of Quantum Materials by Probabilistic Reinforcement Learning
• A Quantum-Inspired Probabilistic Model for the Inverse Design of Meta-Structures
• Quantitative Prediction on the Enantioselectivity of Multiple Chiral Iodoarene Scaffolds Based on Whole Geometry
• Landscape of Big Medical Data: A Pragmatic Survey on Prioritized Tasks
• Value-Added Chemical Discovery Using Reinforcement Learning
• Doubly Bayesian Optimization
• Transformative Machine Learning