Learning symmetry-aware atom mapping in chemical reactions through deep graph matching

Maryam Astero,Juho Rousu

DOI: https://doi.org/10.1186/s13321-024-00841-0

2024-04-24

Journal of Cheminformatics

Abstract:Accurate atom mapping, which establishes correspondences between atoms in reactants and products, is a crucial step in analyzing chemical reactions. In this paper, we present a novel end-to-end approach that formulates the atom mapping problem as a deep graph matching task. Our proposed model, AMNet (Atom Matching Network), utilizes molecular graph representations and employs various atom and bond features using graph neural networks to capture the intricate structural characteristics of molecules, ensuring precise atom correspondence predictions. Notably, AMNet incorporates the consideration of molecule symmetry, enhancing accuracy while simultaneously reducing computational complexity. The integration of the Weisfeiler-Lehman isomorphism test for symmetry identification refines the model's predictions. Furthermore, our model maps the entire atom set in a chemical reaction, offering a comprehensive approach beyond focusing solely on the main molecules in reactions. We evaluated AMNet's performance on a subset of USPTO reaction datasets, addressing various tasks, including assessing the impact of molecular symmetry identification, understanding the influence of feature selection on AMNet performance, and comparing its performance with the state-of-the-art method. The result reveals an average accuracy of 97.3% on mapped atoms, with 99.7% of reactions correctly mapped when the correct mapped atom is within the top 10 predicted atoms.

chemistry, multidisciplinary,computer science, interdisciplinary applications, information systems

What problem does this paper attempt to address?

The paper aims to address the problem of atom mapping in chemical reactions. Specifically, the study proposes a novel end-to-end approach to establish precise atom correspondences between reactants and products through deep graph matching techniques. The main contributions of the paper include: 1. **Proposing an atom mapping model (AMNet) based on deep graph matching**: Utilizing molecular graph representations to capture molecular structural features and processing these graph data through Graph Neural Networks (GNN) to achieve efficient and accurate atom mapping. 2. **Enhancing atom mapping accuracy through symmetry detection**: Introducing the Weisfeiler-Lehman test to identify molecular symmetry, reducing the number of possible mappings and improving the accuracy and efficiency of atom mapping. 3. **Comprehensively mapping all atoms in the entire chemical reaction**: Unlike traditional methods that only focus on major components, the new model can map all atoms in both reactants and products, providing a more complete perspective. 4. **Reducing computational complexity**: By integrating efficient graph matching techniques and symmetry consideration strategies, the model reduces the computational complexity issues common in traditional atom mapping methods. Experimental results show that AMNet achieves an average accuracy of 97.3% on the USPTO reaction dataset, and 99.7% of reactions are correctly mapped when the correctly mapped atoms are within the top 10 predicted atoms. This demonstrates the superior performance of the model in the atom mapping task.

Learning symmetry-aware atom mapping in chemical reactions through deep graph matching

Precise atom-to-atom mapping for organic reactions via human-in-the-loop machine learning

ComABAN: refining molecular representation with the graph attention mechanism to accelerate drug discovery

Quantum Mechanics and Machine Learning Synergies: Graph Attention Neural Networks to Predict Chemical Reactivity

Enumeration Approach to Atom-to-Atom Mapping Accelerated by Ising Computing

A Critical Evaluation of Automatic Atom Mapping Algorithms and Tools

ABC-Net: a Divide-and-conquer Based Deep Learning Architecture for SMILES Recognition from Molecular Images.

A System for Learning Atoms Based on Long Short-Term Memory Recurrent Neural Networks

Highly accurate prediction of reaction performance and enantioselectivity with a uniform machine learning protocol

Learning Chemical Reaction Representation with Reactant-Product Alignment

3DReact: Geometric deep learning for chemical reactions

Substructure-Atom Cross Attention for Molecular Representation Learning

Atomas: Hierarchical Alignment on Molecule-Text for Unified Molecule Understanding and Generation

Mapping the space of chemical reactions using attention-based neural networks

Enhancing Chemical Reaction Monitoring with a Deep Learning Model for NMR Spectra Image Matching to Target Compounds

3DMol-Net: Learn 3D Molecular Representation using Adaptive Graph Convolutional Network Based on Rotation Invariance

Transferability of Atom-Based Neural Networks

AutoTemplate: Enhancing Chemical Reaction Datasets for Machine Learning Applications in Organic Chemistry

Completing and Balancing Database Excerpted Chemical Reactions with a Hybrid Mechanistic-Machine Learning Approach

Atom-Based Machine Learning for Estimating Nucleophilicity and Electrophilicity with Applications to Retrosynthesis and Chemical Stability

A generalized-template-based graph neural network for accurate organic reactivity prediction