QSAR model for pesticide toxicity in bioluminescent Vibrio qinghaiensis sp.-Q67

Abstract

Quantitative structure-activity relationship (QSAR) is a computational approach that correlates chemical structural features with biological activity, enabling toxicity prediction without prior knowledge of nonspecific toxicological modes of action. Despite its potential, the application of QSAR models to predict pesticide toxicity in Vibrio qinghaiensis sp.-Q67 (Q67), a photobacterium, remains underexplored. In this study, we developed QSAR models to assess the toxicity of 11 pesticides detected in the dietary pesticide residue exposure of the Chinese population toward Q67. Molecular descriptors were computed using DRAGON 6.0, followed by principal component analysis and descriptor screening via a variable selection method incorporating leave-one-out (LOO) cross-validation (VIPLS). A robust 7-descriptor QSAR model was established and internally and externally validated using 41 toxicity datasets by Y randomization and cross-validation, respectively. The model demonstrated stable predictive performance for the acute toxicity of the 11 pesticides on Q67. Based on the similarity calculation between the predicted compounds and the training set compounds (K nearest neighbor method), the toxicity predictions generated by our QSAR model are within the defined domain of application. Additionally, we summarized the toxicity of these pesticides using pEC₅₀ as the toxicity index, revealing a toxicity range of 2.88-6.66 μg/L. The toxic effects exhibited either time-dependent or time-independent patterns. By comparing the QSAR descriptors across different organisms, including Q67 bacteria, it was revealed that pesticide toxicity to aquatic organisms is fundamentally linked to the electronic polarization and van der Waals forces of the pesticides.

Keywords: Acute toxicity; Pesticides; QSAR; VIPLS.