纳米金属氧化物对人体细胞毒性效应的构效关系研究  

Research on the structure-activity relationship of the cytotoxicity of metal oxide nanoparticles on human cells

作　　者：曹洪印[1,2] 袁蓓蕾 潘勇 史海华[1] CAO Hong-yin;YUAN Bei-lei;PAN Yong;SHI Hai-hua(Jiangsu Key Laboratory of Hazardous Chemicals Safety and Control,Nanjing Tech University,Nanjing 210009,China;School of Environmental Engineering,Nanjing Polytechnic Institute,Nanjing 210048,China)

机构地区：[1]南京工业大学江苏省危险化学品安全与控制重点实验室,南京210009 [2]南京科技职业学院环境工程学院,南京210048

出　　处：《安全与环境学报》2023年第1期304-312,共9页Journal of Safety and Environment

基　　金：国家自然科学基金项目(81803274)。

摘　　要：为建立高效的纳米金属氧化物细胞生物毒性构效关系预测模型,研究了20种纳米金属氧化物在不同生物条件下对人正常肺上皮细胞(BEAS-2B)和角质层细胞(HaCaT)的毒性效应构效关系,并首次将元素周期描述符(定量描述符)与试验条件参数(定性描述符)相结合,共同表征金属氧化物的纳米结构特征。在采用支持向量机-特征递归消除法(Support Vector Machine-Recursive Feature Elimination,SVM-RFE)筛选的最优描述符作为输入参数的基础上,分别应用支持向量机(Support Vector Machine,SVM)和随机森林(Random Forest,RF)2种高效的建模方法,建立纳米材料构效关系(Structure-Activity Relationships for Nanoparticals,Nano-SAR)预测模型。2个算法训练集的准确率(ACC)均大于0.9,内部验证准确率均大于0.7,测试集外部验证的准确率也均大于0.8,模型验证结果表明2个算法均具有良好的稳定性和较强的预测能力。对比2个算法研究结果表明,RF算法优于SVM算法,且优于文献报道的已有算法。模型机理解释结果表明,水合粒径和电负性是影响两种细胞毒性的主要因素。To establish an efficient structure-activity relationship model for cytotoxicity of metal oxide nanoparticles,the structureactivity relationships of toxic effects of twenty metal oxide nanoparticles on human normal lung epithelial cells(BEAS-2B)and keratinocytes(Hacat)were researched under different biological conditions.For the first time,eleven periodic tablebased descriptors(quantitative descriptors)and six experimental condition parameters(qualitative descriptors)were combined to characterize the nanostructure characteristics of metal oxides.Three descriptors,namely metal electronegativity(χ),Hydro Size and Dose,were selected from the above descriptors by the support vector machine-feature recursive elimination(SVM-RFE)method to form the optimal feature subset,which was used as input parameters of this study.On this basis,two nano-SAR prediction models were established by using support vector machine(SVM)and random forest(RF)modeling methods respectively.Both accuracy(ACC)of training sets in two models are more than 0.9,both accuracies of the internal validation are more than 0.7,and both accuracies of the external validation for the test sets are more than 0.8.The model validation results show that the combination of periodic tablebased descriptors and biological condition parameters can effectively characterize the molecular structure characteristics of metal oxide nanoparticles,and the two models established have good stability and strong prediction ability.The results of the model comparison show that the RF model is superior to the SVM model,and the performance of the RF model is superior to existing models reported in the literature.The results of model mechanism interpretation using the descriptor sensitivity analysis method show that the Hydro Size and electronegativity of metal oxide nanoparticles are the main structural factors affecting their toxicity to both human normal lung epithelial cells(BEAS-2B)and keratinocytes(Hacat),and the smaller the nanoparticle size is,the easier it is to enter the c

关 键 词：环境科学技术基础学科 纳米金属氧化物 生物毒性 构效关系 元素周期描述符 随机森林 模型 

分 类 号：X912[环境科学与工程—安全科学]