doi:10.16597/j.cnki.issn.1002-154x.2026.02.010
分子气味预测的定量构效关系解析：数据、模型与跨学科挑战
李彦鹏1 李晓平2 盛世杰2 刘玉坤2 朱亚峰2 周建成1 李乃旭1*
（1.bet体育365官网网站，南京211189；2.江苏鑫源烟草薄片有限公司，淮安223002）
摘 要 气味感知是生物体解析化学信号的关键过程之一，其本质是气味分子与嗅觉受体的特异性结合引发的神经激活过程。然而，分子结构与其引发的嗅觉感知之间的非线性映射关系尚未完全阐明，传统人工嗅辨存在主观性、低通量缺陷。借助机器学习开展定量构效关系建模是气味智能预测的有效路径。作者系统梳理相关研究进展，从数据库资源、分子特征表征、建模算法三大维度展开论述，对比各类数据库与化学信息学工具，剖析线性模型、SVM、图神经网络等模型优劣，论述多学科融合难点，展望受体-分子多级联合建模发展前景。
关键词 机器学习 定量构效关系 分子气味预测 化学信息学
Deciphering Quantitative Structure-Activity Relationships for Molecular Odor Prediction: Data, Models and Interdisciplinary Challenges
LI Yanpeng1,LI Xiaoping2,SHIJie2,LIU Yukun2,ZHU Yafeng2,ZHOU Jiancheng1,LI Naixu1*
(1.School of Chemistry and Chemical Engineering,Southeast University,Nanjing 211189,China;2.Jiangsu Xinyuan Tobacco Sheet Co.,Huaian 223002,China)
Abstract Odor perception, a critical process for organisms to decode chemical signals, involves the specific binding of odor molecules to olfactory receptors, triggering neural activation. However, the nonlinear mapping between molecular structures and resulting olfactory perceptions remains incompletely elucidated. Traditional sensory evaluation methods face limitations due to subjectivity and low throughput, hindering progress in this field. Machine learning has emerged as a powerful tool to address these challenges by modeling quantitative structure-activity relationships. This review systematically reviews progress in molecular odor prediction using quantitative structure-activity relationships approaches, focusing on three core technical dimensions: data resources, molecular representation modeling methods, and interdisciplinary integration strategies. Future prospects are discussed for integrating molecular odor prediction with quantum chemistry calculations, single-cell sequencing, and interpretable artificial neural networks to enhance model transparency and biological relevance.
Keywords machine learning quantitative structure-activity relationship molecular odor prediction cheminformatics tools