Development of an Interpretable QSAR Model for Predicting Acute Oral Toxicity of Organophosphates in Rats Based on GA-MLR Algorithm.

Abstract

Organophosphates (OPs) are highly hazardous chemicals with broad-spectrum toxicity. Traditional in vivo methods for determining OP toxicity are time-consuming and labor-intensive. In this study, we developed a quantitative structure-activity relationship (QSAR) model to predict acute rat toxicity of OPs using two-dimensional molecular and quantum chemical descriptors, optimized through genetic algorithm-based multiple linear regression (GA-MLR). The optimal model demonstrated robust performance with the following statistical parameters: coefficient of determination (R²) of 0.7451, leave-one-out cross-validation (LOOCV) coefficient (Q²_Loo) of 0.6208, external test set coefficient of determination (R²_ext) of 0.7360. These metrics indicate excellent generalization and predictive capabilities of the model. Interpretative analysis of the model revealed that NumHDonors and PEOE_VSA were the most significant descriptors influencing OP toxicity. An increase in hydrogen bond donors within OP molecules reduces toxicity, as these donors enhance hydrophilicity, diminishing membrane permeability. Moreover, the PEOE_VSA descriptor characterizes the partial charge properties of OP molecules, reflecting their electrostatic interactions with acetylcholinesterase (AChE) during binding, which influences toxicity. This study presents an optimized modeling strategy designed for small datasets, enabling stable feature selection and accurate assessment of their contributions to toxicity prediction. This research provides a reliable QSAR approach for OP toxicity prediction while offering new insights into toxicity mechanisms.