Development and Validation of a Multiple Linear Regression QSAR Model for the Prediction of Acute Aquatic Toxicity (LC₅₀) of Herbicides

The quantitative structure-activity relationship (QSAR) approach constitutes a powerful computational framework for establishing systematic and quantitative correlations between molecular structure and measurable biological or toxicological endpoints. By integrating the strategic selection of molecular descriptors, encoding critical topological, electronic, and physicochemical features, with rigorous chemometric workflows and multivariate statistical modeling, this methodology delivers high predictive accuracy, internal stability, and external generalizability. In the present study, a QSAR model based on Multiple Linear Regression (MLR) was developed and rigorously validated for the prediction of acute aquatic toxicity, expressed as LC50, of a structurally diverse series of herbicides. A curated dataset of 110 compounds, sourced from peer-reviewed literature, was employed for model construction and validation. Dataset partitioning into training and external test sets was performed using the Kennard-Stone algorithm, ensuring representative and uniform coverage of the chemical descriptor space. Model validation was conducted in full compliance with the Organisation for Economic Co-operation and Development (OECD) guidelines for QSAR model validation, encompassing internal validation via leave-one-out cross- validation, external validation on an independent test set, Y-randomization testing to exclude chance correlations, and applicability domain assessment through the Williams plot approach. The developed model demonstrated satisfactory and reliable statistical performance in terms of goodness-of-fit, robustness, and external predictive capability, meeting all established acceptance criteria. The mechanistic interpretation of the selected molecular descriptors provided meaningful and chemically coherent insights into the structural and physicochemical determinants governing herbicide acute toxicity. The proposed model offers a valuable, transparent, and cost-effective in silico tool for environmental risk assessment, regulatory decision-making, and the rational design of novel agrochemicals with optimized efficacy and reduced ecological impact.