摘要
纳米金属氧化物(MO_(x))易通过呼吸进入肺部,诱发肺细胞毒性,进而导致肺部多种疾病。仅通过实验方法评估毒性面临效率低、成本高和伦理问题等局限性。对MO_(x)的肺部细胞毒性进行全面评估,有利于防治新污染物的健康危害。为了充分利用现有毒性数据,本研究开发了肺细胞(A549细胞)毒性高通量预测模型,对传统毒性实验进行了补充。通过文献挖掘,构建了MO_(x)诱导肺细胞毒性数据集,涵盖9种MO_(x)在8种不同浓度下产生的72个毒性数据点,以及材料密度、粒径等29种性质参数。使用合成少数类过采样技术(SMOTE)算法处理不平衡数据,构建了基于多种机器学习算法的预测模型,表现最优的模型在训练集十折交叉验证和测试集外部验证中准确率分别超过0.90和0.85。基于Shapley可加性特征解释方法(SHAP),识别了暴露浓度、暴露时间以及材料分散指数等影响毒性的5个关键参数,阐明了毒性机理。基于相似性网络图对建模数据的代表性及模型应用域进行了表征。本研究构建的机器学习模型实现了MO_(x)毒性的高通量准确预测,可以作为纳米材料毒性评估和安全设计的有力工具。
Metal oxide nanoparticles(MO_(x))can easily enter the lungs through respiration,inducing cytotoxicity in lung cells and subsequently leading to various lung diseases.To address the health hazards of new pollutants,a comprehensive assessment of the toxicity of MO_(x)is crucial.However,evaluating toxicity solely through experimental methods faces limitations such as low efficiency,high costs,and ethical concerns.To fully utilize the existing toxicity data,this study developed a high-throughput predictive model for lung cell toxicity(A549 cells)to complement traditional toxicity experiments.Through literature mining,a dataset was constructed that included 72 toxicity data points induced by 9 types of MO_(x)at 8 different concentrations,along with 29 property parameters such as material density and particle size.The Synthetic Minority Over Sampling Technique(SMOTE)was employed to handle imbalanced data,and predictive models were built using various machine learning algorithms,achieving a prediction accuracy exceeding 0.95 in ten-fold cross-validation on the training set and over 0.85 in external validation on the test set.Using the Shapley Additive Explanations(SHAP)method,five key parameters influencing toxicity were identified,including exposure concentration,exposure time,and material dispersion index,which elucidated the toxicity mechanisms.The representativeness of the modeling data and the applicability domain of the model were characterized using similarity network graphs.The machine learning model established in this study enables high-throughput and accurate predictions of nanotoxicity,serving as a powerful tool for the toxicity assessment and safe design of nanomaterials.
作者
王天勤
郭丰林
孟伽娜
张子琦
李建青
孙浩
徐文龙
张洪武
黄杨
李斐
WANG Tianqin;GUO Fenglin;MENG Jiana;ZHANG Ziqi;LI Jianqing;SUN Hao;XU Wenlong;ZHANG Hongwu;HUANG Yang;LI Fei(College of Chemistry and Materials Science,Ludong University,Yantai 264025,China;Key Laboratory of Coastal Environmental Processes and Ecological Restoration,Chinese Academy of Sciences(Yantai Institute of Coastal Research),Key Laboratory of Coastal Environmental Processes of Shandong Province,Yantai Institute of Coastal Research,Chinese Academy of Sciences,Yantai 264003,China)
出处
《生态毒理学报》
北大核心
2025年第1期23-35,共13页
Asian Journal of Ecotoxicology
基金
国家自然科学基金青年项目(22406080)
国家自然科学基金面上项目(22376215,U22A20618)
山东省自然科学基金项目(ZR2023MD071)
泰山学者工程(tsqn202312275)
山东省高等学校优秀青年创新团队项目(2023KJ213)。
关键词
纳米颗粒
预测模型
机器学习
细胞毒性
肺部毒性
nanoparticle
predictive model
machine learning
cytotoxicity
lung toxicity
