基于极限梯度提升算法和特征筛选方法的羊角月牙藻(Selenastrum capricornutum)急性毒性定量构效关系(QSAR)模型的建立与应用  

Development and Application of Quantitative Structure-activity Relationship(QSAR)Model for Acute Toxicity of Selenastrum capricornutum Based on Extreme Gradient Boosting Algorithm and Feature Selection Method

作　　者：滕跃发 王晓晴 李斐 吉成龙[1,3] 吴惠丰[1,3] Teng Yuefa;Wang Xiaoqing;Li Fei;Ji Chenglong;Wu Huifeng(CAS Key Laboratory of Coastal Environmental Processes and Ecological Remediation,Yantai Institute of Coastal Zone Research,Chinese Academy of Sciences,Shandong Key Laboratory of Coastal Environmental Processes,Yantai Institute of Coastal Zone Research,Chinese Academy of Sciences,Yantai 264003,China;University of Chinese Academy of Sciences,Beijing 100049,China;Center for Ocean Mega-Science,Chinese Academy of Sciences,Qingdao 266071,China)

机构地区：[1]中国科学院海岸带环境过程与生态修复重点实验室(烟台海岸带研究所),山东省海岸带环境过程重点实验室,中国科学院烟台海岸带研究所,烟台264003 [2]中国科学院大学,北京100049 [3]中国科学院海洋大科学中心,青岛266071

出　　处：《生态毒理学报》2023年第3期33-46,共14页Asian Journal of Ecotoxicology

基　　金：烟台市科技创新发展计划项目(2020MSGY060);国家自然科学基金资助项目(21677173)。

摘　　要：藻类是水生食物网中主要的初级生产者,对水生生态系统的可持续性起着重要作用。随着社会发展、工业进步和人类活动,大量化学品被释放到水生环境中,对藻类产生了极大的威胁。若藻类受到危害势必会影响其他水生生物,因此有必要开展藻类的毒性评估。藻类的毒性评估需要大量的毒性数据,通过实验的方法获得水生毒性数据成本较高且比较耗时,定量构效关系(QSAR)是解决这类问题的一种良好的替代方法。本研究基于Web of Science与中国知网数据库文献中的53条急性毒性数据,利用极限梯度提升(XGB)算法和特征筛选方法建立了羊角月牙藻(Selenastrum capricornutum)急性毒性的QSAR模型。最优模型的训练集决定系数(R^(2)_(TR))达到了0.97,验证集决定系数(Q^(2)_(EXT))达到了0.78,留一法交叉验证系数(Q^(2)_(LOO))也达到了0.51,表明建立的QSAR模型具有较好的拟合优度、稳健性和预测能力。机理解释结果表明,化合物的拓扑电荷数、总原子序数和电负性是影响羊角月牙藻急性毒性的关键因素。在此基础上,采用建立的QSAR模型和EPI Suite分别预测了16种典型多环芳烃(PAHs)对藻类的急性毒性,并对其进行了毒性分级。研究结果为藻类的急性毒性数据的获取提供了一个高效预测工具,有利于加快化学品的水环境风险评估工作。Algae as the main primary producers in aquatic food webs play an important role in ensuring the sustainability of aquatic ecosystems.However,a large number of chemicals have been released into the aquatic environment with the development of industrial production and countless other human activities,posing a great threat to algae.If algae are endangered,they will inevitably affect other aquatic organisms.Therefore,it is imperative to assess environmental toxicity on algae.The assessment requires a large amount of toxicity data through experimental measurements,which is costly and time consuming.Quantitative structure activity relationships(QSAR)is a good alternative method to solve these problems.In this study,QSAR models for the acute toxicity of Selenastrum capricornutum were constructed by using the extreme gradient boosting(XGB)algorithm and feature selection method.53 acute toxicity data were gathered from Web of Science and China National Knowledge Infrastructure.The optimal model achieved a coefficient of determination(R^(2)_(TR))of 0.97 for training set,a coefficient of determination(Q^(2)_(EXT))of 0.78 for validation set,and a leave-one-out cross-validation coefficient(Q^(2)_(LOO))of 0.51,respectively.In addition,the results showed that the topological charge number,total atomic number and electronegativity of the compounds were the key factors affecting the acute toxicity of Selenastrum capricornutum.On this basis,the established QSAR model and EPI Suite were used to predict the acute toxicity of 16 typical polycyclic aromatic hydrocarbons(PAHs)to algae,respectively.This study provides an efficient predictive tool for obtaining acute toxicity data of algae and helps to accelerate environmental risk assessment of algae.

关 键 词：多环芳烃 羊角月牙藻 定量构效关系 机器学习 极限梯度提升 特征筛选 

分 类 号：X171.5[环境科学与工程—环境科学]