Comparative evaluation of mathematical models for predicting acute toxicity of chemicals

Abstract

Introduction. Considerable attention is paid to the assessment of acute toxicity of chemical compounds during oral administration due to the different rates of absorption of substances in different animal species and various experimental conditions. Given the pace of development of the chemical industry, researchers are faced with the question of accelerating the study of the properties of substances and filling data gaps. Therefore, quantitative prediction of the toxic properties of substances using mathematical models based on the structure or structural properties of compounds - quantitative structure -activity relationship (QSAR) modeling - is one of the promising areas. The purpose of this study is to create and compare the performance of the obtained mathematical models for predicting the acute toxicity of various classes of chemicals. Materials and methods. The study included four classes of pesticides (organochlorine compounds (OCs), azoles, carbamates, organophosphorus compounds (OPs) in the amount of 100 compounds with descriptors calculated by PaDEL-Descriptors software ver. 2.21. Regression models were constructed in the WEKA software, subjected to an internal validation procedure. Statistical parameters such as the mean square error (RMSE) and the coefficient of determination (r 2) were used to assess the quality of regression models. Results. To predict acute oral toxicity of OCs and OPs, it is optimal to use a model in which neural networks and the support vector method are combined, for carbamates - an ensemble model that includes linear regression and the support vector method. For substances from the azole group, it was not possible to create a model that would meet the necessary requirements: r 2>0.6 for the training set and r 2 >0.5 for cross-validation. Limitations. The study is limited by the number of compounds studied, the class of chemical compounds, and the area of distribution of the results obtained during modeling. Conclusion. In this study, ensemble modelling methods demonstrated the best results in predicting acute oral toxicity for OCs, carbamates, and OPs.