Development of QSAR models for evaluating pesticide toxicity against Skeletonema costatum

Abstract

Nowadays, the emergence of pesticides and its application in agriculture greatly improved the crop quality and food production. However, the resulted ecological problem caused by the widespread pesticide residues attracted more and more attention since the pesticides were harmful to most living organisms. Regulatory agencies such as Environmental Protection Agency (EPA) and European Chemicals Agency (ECHA) stipulated that a comprehensive pesticides risk assessment was essential and also underscored the application of computation method in evaluating pesticides. The present study aimed to use the Quantitative Structure-Activity Relationship (QSAR) method to establish models for quantitatively and qualitatively predicting the toxicity of pesticide against Skeletonema costatum. The regression model was developed using the Genetic Algorithm plus Multiple Linear Regression method and the classification model was established based on the Random Forest algorithm, respectively. Various internal and external validation metrics suggested that the obtained regression model was of good fitness (R2=0.722), robustness (QLOO2=0.653) and external predictive ability (QFn2:0.719–0.776, CCC = 0.878). The classification could correctly predict 79.4% of pesticides in the training set and 69.7% in the validation set. The relatively high sensitivity value of the classification model indicated its good performance in identifying high-toxic pesticides. It could be concluded from the selected modelling descriptors that molecular weight and polarizability impacted the toxicity the most. The atom-type E-state descriptors generally contributed negatively to the pesticide toxicity which verified the negative influence of molecular hydrophilicity. Moreover, the lipophilic, carbon-type, charge related descriptors demonstrated the important influence of lipophilicity and polarity on pesticide toxicity. The models presented in this work could be used to pre-evaluate the toxicity of pesticides within the applicability domain, thus focusing resources on the high-toxic pesticides and assessing the environmental risk of pesticides quickly and economically.