摘要
为满足大量化学品危险特性的评价需要,定量结构-性质关系(QSPR)在化学品安全领域的研究与应用得到了迅猛发展。QSPR研究的发展侧重模型预测能力或注重机理解释,都要依靠计算方法的不断优化来提高模型的拟合能力、预测能力、稳定性或处理复杂混合物的能力等。本文综述了QSPR研究在化学品危险特性评价中的发展,特别是特征描述符选择算法、建模算法和基于图神经网络的深度QSPR的技术原理与应用实践。
In order to meet the need of evaluating the hazardous characteristics of a large number of chemicals,the research and application of quantitative structural-property relationship(QSPR)in the field of chemical safety have been developed rapidly.The research of QSPR focuses on the prediction ability of model or the mechanism explanation,both of which rely on the continuous optimization of calculation methods to improve the fitting ability,prediction ability,stability or the ability to deal with complex mixtures.This paper reviews the development of QSPR research in the evaluation of chemical hazard characteristics,especially the technical principle and application practice of feature descriptor selection algorithm,modeling algorithm and deep QSPR based on graph neural network.
作者
马倩
宋项宁
苑媛
万可风
杜红岩
张宏哲
郭亚逢
张金梅
MA Qian;SONG Xiang-ning;YUAN Yuan;WAN Ke-feng;DU Hong-yan;ZHANG Hong-zhe;GUO Ya-feng;ZHANG Jin-mei(State Key Laboratory of Chemical Safety,SINOPEC Research Institute of Safety Engineering Co.,Ltd.,Qingdao 266000,China;National Registration Center for Chemicals,Ministry of Emergency Management,Qingdao 266000,China;SINOPEC Health Safety and Environmental Management Department,Beijing 100728,China)
出处
《化学研究与应用》
北大核心
2025年第3期513-523,共11页
Chemical Research and Application
基金
中国石化科技部项目(H24009)资助。
关键词
定量结构-性质关系
化学品安全
特征选择
机器学习
图神经网络
quantitative structure-property relationship
chemical safety
feature selection
machine learning
graph neural network
