Quantitative Structure Activity Relationship Prediction Of The Toxicity Of Polycyclic Aromatic Hydrocarbons In Atmospheric Particles

PAHs is a great harmful to the environment and human health, and an atmospheric organic pollutant with carcinogenic, teratogenic and mutagenic effect. With the understanding of the effect of toxicity of polycyclic aromatic hydrocarbons(PAHs), the toxic mechanism of polycyclic aromatic hydrocarbons(PAHs) will be regarded as an important topic of security field to research by the researchers at home and abroad. is one of the most effective methods to detect the organic pollutants toxicity in the current world. The toxicity is obtained from the molecule structure directly by the mathematical model, which avoids the expensive cost of traditional experimental methods, complex process and other defects. Hence developing QSAR of PAHs toxicity is significant to predict and evaluate ecological risk of PAHs, control pollution and prevent.The toxicity of polycyclic aromatic hydrocarbons(PAHs) is studied by this paper, with developing the carcinogenicity of polycyclic aromatic hydrocarbons and quantitative relation(QSAR) prediction of air- octanol partition coefficient(KOA) mainly. What using the support vector machine(SVM) algorithm establishes the model of structure-activity, is proposed by the nonlinear characteristics of PAHs with less sample data. What using the grid search method(GSM), genetic algorithm(GA) and Particle Swarm Optimization(PSO) optimizes the parameter, is proposed by the problem that the parameter selection has a great influence on the support vector machine(SVM). Specific studies are as follows:1 At first, the data was normalized and the model was established by using support vector regression(SVR) algorithm about the quantitative structure activity relation(QSAR) prediction of KOA regression of polycyclic aromatic hydrocarbons(PAHs). The fitting decision coefficient and mean squared error(MSE) of the regression model are calculated and the prediction ability is verified by the result of the artificial neural network(ANN) algorithm. At second, the model is optimized by the grid search method(GSM), genetic algorithm(GA) and Particle Swarm Optimization(PSO). Eventually fitting the decision coefficient of the support vector machine is more than the artificial neural network(ANN) and the mean squared error(MSE) is less than that of the artificial neural network(ANN). The grid search method(GSM) and Particle Swarm Optimization(PSO) algorithm are better than the original model. The optimization of the grid search method(GSM) exists fitting phenomenon and the effect of prediction is not good. In comprehensive analysis, the grid search optimization model is the best model of regression prediction.2 At first, the data is on dimensionality reduction by principal component analysis method and establishes the implementation mode by using support vector machine(SVM) classification algorithm about the quantitative structure activity relation(QSAR) prediction of the PAHs carcinogenicity classification. The classification accuracy of classification model is calculated and prediction ability is verified by the result of the artificial neural network(ANN) algorithm. At second, the parameter is selected by using the grid search method(GSM), genetic algorithm(GA) and Particle Swarm Optimization(PSO). The model is established with the best combination of parameters. Eventually the classification accuracy of classification prediction in the support vector machine(SVM) is more than that of the artificial neural network(ANN). The model prediction results of the grid search method(GSM) and Particle Swarm Optimization(PSO) algorithm are better than the original model. In comprehensive analysis, the genetic optimization model is the best classification prediction model.