有机化学    

综述与进展

人工智能驱动的化学反应机理研究进展

梁鹏宇, 王陈辰, 姚小军*   

  1. 澳门理工大学 应用科学学院 人工智能药物发现中心 澳门 999078
  • 收稿日期:2026-07-14 修回日期:2026-08-24
  • 基金资助:
    澳门科学技术发展基金(No. 0030/2024/RIA1)和澳门理工大学研究基金(No. RP/FCA-15/2023)资助项目.

Advances in AI-driven Chemical Reaction Mechanism Research

Liang, Pengyu, Wang, Chenchen, Yao, Xiaojun*   

  1. Center for Artificial Intelligence Driven Drug Discovery, Faculty of Applied Sciences, Macao Polytechnic University, Macao, 999078
  • Received:2026-07-14 Revised:2026-08-24
  • Contact: *E-mail: xjyao@mpu.edu.mo
  • Supported by:
    Science and Technology Development Fund, Macau SAR (No. 0030/2024/RIA1), and Macao Polytechnic University (No. RP/FCA-15/2023).

化学反应机理是明确反应过程、解释反应规律并指导反应设计的重要基础。传统反应机理研究通常利用实验表征捕获反应中间体、获取动力学和产物分布等信息,并结合理论计算评估候选反应路径、过渡态结构和反应能垒。然而,面对多构象、多组分等复杂反应体系,机理假设的提出和验证仍依赖人工经验,同时受到计算成本和搜索空间规模的限制。人工智能能够从分子结构、反应数据和量子化学计算结果中学习化学规律,在复杂反应空间的高效探索和机理研究的自动化中展现出巨大潜力。本文首先概述了化学结构的主要表征方法及其适用模型。在此基础上,围绕化学反应机理研究的关键环节,系统综述了人工智能在反应位点识别、反应机理推断、过渡态生成、反应能垒与选择性评价,以及机器学习势与反应势能面构建中的研究进展。最后,本文总结了现有方法在数据质量、物理化学约束、模型泛化能力和可解释性等方面面临的问题,并讨论了其与实验表征和理论计算协同发展的可能方向。

关键词: 人工智能, 化学反应机理, 分子表征, 基元反应, 过渡态, 机器学习势

Elucidating chemical reaction mechanisms is essential for clarifying reaction processes, rationalizing reactivity patterns and guiding reaction design. Traditional mechanistic studies use experimental characterization to detect reaction intermediates and obtain kinetic information and product distributions, while theoretical calculations are employed to evaluate candidate reaction pathways, transition-state structures and reaction barriers. However, for complex reaction systems involving multiple conformations and reaction components, the proposal and validation of mechanistic hypotheses still rely heavily on human expertise, and are limited by high computational costs and the large scale of the search space. Artificial intelligence can learn chemical patterns from molecular structures, reaction data, and quantum chemical calculation results, and demonstrate considerable potential for efficiently exploring complex reaction spaces and advancing the automation of mechanistic studies. This review first summarizes the main representations of chemical structures and their applicable model architectures. On this basis, recent advances in artificial intelligence for key tasks in chemical reaction mechanism research are systematically reviewed, including reaction-site identification, reaction mechanism inference, transition-state generation, reaction barrier and selectivity prediction, and machine-learning potentials for reaction potential energy surface construction. Finally, current challenges related to data quality, physicochemical constraints, model generalization, and interpretability are summarized, and possible directions for the integration of artificial intelligence with experimental characterization and theoretical calculations are discussed.

Key words: artificial intelligence, chemical reaction mechanism, molecular representation, elementary reaction, transition state, machine-learning potential