数据驱动的有机分子理化性质预测  

Data-driven prediction of physicochemical properties oforganic molecules

作　　者：孙一舟 汤缪炅 张硕卿 洪鑫 Yi-Zhou Sun;Miaojiong Tang;Shuoqing Zhang;Xin Hong(Center of Chemistry for Frontier Technologies,Department of Chemistry,Zhejiang University,Hangzhou 310027,China;School of Chemistry and Chemical Engineering,Henan Normal University,Xinxiang 453007,China)

机构地区：[1]浙江大学化学系,浙江大学化学前瞻技术研究中心,杭州310027 [2]河南师范大学化学化工学院,新乡453007

出　　处：《科学通报》2025年第4期492-507,共16页Chinese Science Bulletin

基　　金：国家重点研发计划(2022YFA1504301);国家自然科学基金(22122109,22271253,22103070);浙江省自然科学基金(LDQ23B020002);浙江大学上海高等研究院繁星科学基金(SN-ZJU-SIAS-006);中国科学院青年交叉团队(JCTD-2021-11);中央高校基本科研业务费专项资金(226-2022-00140,226-2022-00224,226-2023-00115,226-2024-00003);固体表面物理化学国家重点实验室项目(202210);浙江省科技厅领军型创新创业团队项目(2022R01005);河南师范大学化学化工学院开放研究基金(2024Z01)资助。

摘　　要：分子的理化性质对于化学行为具有决定性影响,其精确高效预测是化学和材料科学领域研究的长期热点之一.随着理化性质数据的积累和分子人工智能(AI)建模方法的进步,数据驱动的分子理化性质预测迎来了跨越式发展,在精度、广度、效率上取得了一系列突破.在有机分子理化性质的AI预测方面,本文介绍了相关数据库的发展,围绕光谱性质、轨道能量、酸度系数(pKa)、键解离能(BDE)、亲核性参数等代表性场景探讨了机器学习预测的相关进展,并对该领域的现状进行了总结与展望.The physicochemical properties of molecules, such as frontier orbital energy levels, bond energies, and spectroscopiccharacteristics, form the basis for understanding and predicting molecular chemical behavior. For instance, based on theenergy levels and arrangement of frontier orbitals, chemists can effectively determine electron donors and acceptors inchemical reactions, and rationally assess reaction activity and selectivity, widely applied in a series of organic moleculartransformations including pericyclic reactions. Chemical bond energies provide core thermodynamic parameters for theprocess of bond breaking and formation, which are crucial for precisely understanding reaction thermodynamics andkinetics. UV, IR, and NMR spectroscopic features offer effective means to describe the geometric and electronic structuresof molecules. Therefore, an accurate understanding of molecular physicochemical properties is one of the importantstrategies for humans to explore the molecular world, which not only helps in deeply understanding the microscopicmechanisms of chemical reactions but also effectively guides the design and development of new reactions, making it along-term focus of chemical scientific research.Traditionally, molecular properties were determined through experimental measurements and theoretical calculations.While experiments provide a reliable foundation, they are often constrained by cost, efficiency, and the need for specializedequipment. Computational chemistry, bolstered by advancements in computer hardware, has emerged as a complementaryapproach, extending the reach of molecular property research through simulations of electronic structure and reactiondynamics. However, computational methods also face challenges in achieving precision and efficiency in complexmolecular systems.With the continuous accumulation of chemical data and significant advancements in artificial intelligence (AI)technology, machine learning methods have made remarkable progress in chemistry, showing important potential in the

关 键 词：分子性质预测 分子数据库 机器学习 AI化学 

分 类 号：O621[理学—有机化学] TP18[理学—化学]