Ecotoxicity Prediction And Risk Assessment Of Organic Compounds Based On Ensemble Machine Learning Methods

With the acceleration of urbanization,the production and use of organic compounds is increasing,bringing many conveniences and benefits to mankind.However,the presence of organic chemicals is accompanied by environmental pollution,affecting all aspects of natural resources,and its toxic and side effects have brought great negative effects on the ecological environment,including the atmosphere,soil,water and dependent organisms.Traditional microorganisms' methods of degrading organic pollutants are cumbersome,occupy a large area,and have high costs.They cannot prevent pollutants from entering the environment from the source,and they are increasingly not adapted to the rapid development of modern society.Machine learning methods have been used to predict and evaluate the ecological toxicity of organic compounds,which can efficiently and quickly determine the potential risks of these chemicals to humans and the ecological environment,and help to carry out in-depth research on the biodegradation of organic compounds.In this study,molecular fingerprints combined with three machine learning methods(including random forest,support vector machine,and extreme gradient boosting)were used to develop integrated classification and prediction models for the ecological toxicity of organic compounds to the atmospheric,terrestrial,and aquatic environments After repeated 5-fold cross-validation,the overall prediction accuracy of the ensemble model on the data set from the three types of model organisms reached 88.7%,76.1% and 92.6%,respectively,and the area under the receiver operating characteristic curve(AUC)are 0.870,0.838 and 0.966.Compared with previously reported methods,the ensemble model achieves higher accuracy and AUC values.Finally,the random forest algorithm was used for feature importance analysis,and some structural characteristics that could represent ecological toxicity were identified.In summary,this paper has the following innovative work:(1)Three ensemble classification models for the prediction of the ecological toxicity of organic compounds have been developed.The models have obtained better performance parameters and provide reliable predictions for the ecological toxicity of organic compounds.Technical support and assessment tools;(2)identified some representative alert structures that represent ecotoxicity,providing strong theoreticalsupport and guidance value for better screening of toxic pollutants and ecological risk assessment.