Study Of Safety Evaluation Model For Exogenous Compounds

ObjectiveTo explore the application of artificial intelligence algorithm in the safety evaluation of exogenous compounds.Taking carcinogenicity and ionic liquid phototoxicity of compounds as the research object,the quantitative structure-activity/property relationship method was developed to construct an efficient prediction model for evaluating the safety of exogenous compounds,and to provide ideas and guidance for the design of new compounds with lower toxicity and environmental friendly compounds.MethodsThe classification prediction model of carcinogenicity of aromatic amine was established by using multi-layer perceived artificial neural network algorithm and gene expression programming algorithm.Models were evaluated by comparing the accuracy,sensitivity,specificity,and Youden index.Physical and chemical factors about the carcinogenicity of aromatic amines were summarized.Heuristic algorithm and gene expression programming algorithm were used to establish a model for predicting the phototoxicity of ionic liquids.Models were evaluated by comparing the square of the correlation coefficient(R~2)and the error(S~2).Physical and chemical factors influenced the phototoxicity of the ionic liquid were analyzed.Results1.The accuracy,sensitivity,specificity,Youden index of training and test sets in the prediction model of carcinogenicity of aromatic amine based on multi-layer perceived artificial neural network algorithm were 0.871,0.640,0.956,0.596 and 0.771,0.556,0.846,0.402,respectively.The accuracy,sensitivity,specificity,Youden index of training and test set in the prediction model of carcinogenicity of aromatic amine based on gene expression programming algorithm were 0.914,0.720,0.985,0.705 and 0.829,0.667,0.885,0.552,respectively.Chemical variables Kier flexibility index,Balaban index,Polarity parameter,the lower the LUMO energy,Number of carbon atoms,Structural information content index-order,Topographic electronic index-all bonds and Number of nitrogen-atoms were in line with the modeling requirements,and they were closely related to the activity of aromatic amines.2.In the phototoxicity prediction model for ionic liquids,the optimal results were obtained basis on five physicochemical variables.In the heuristic method model:training set R~2=0.95,S~2=0.0654;the test set R~2=0.88 and S~2=0.32.In the gene expression programming algorithm model:training set R~2=0.97,S~2=0.04;the test set R~2=0.91,S~2=0.22.In the heuristic algorithm,the Kier shape index(order 3),Number of F atoms,and Avg bond order of a C atom were positive correlation with ionic liquid toxicity.The relative number of double bonds and relative negative charged surface area were negative correlation with ionic liquid toxicity.Conclusion1.Quantitative structure-property relationships are innovative ways to predict the carcinogenicity of aromatic amines.In this study,the predictive ability of the gene expression programming algorithm is better than the multi-layer perceived artificial neural network algorithm.Models by the two methods have higher predictive ability and predictive reliability.The unique advantage of an expression programming algorithm is a good mathematical algorithm for the predictive model.2.Ionic liquid toxicity prediction model by gene expression programming algorithm is superior to heuristic algorithm model.Gene expression programming algorithm has been successfully applied to the prediction of ionic liquid phototoxicity,and a more accurate quantitative structure-activity relationship prediction model has been established,which shows that the nonlinear model may be a better way in the study of ionic liquid toxicity.3.Quantitative structure-activity/property relationship analysis method can be successfully applied to exogenous compounds.Safety evaluation model can achieve rapid prediction of the new compound properties,toxicity,biological activity;and can provide ideas for the design of new green safety compounds.