Generation of explainable QSAR models for predicting chemical-induced urinary tract toxicity

Chemical-induced urinary tract toxicity, particularly in the bladder and ureters, remains undercharacterized relative to nephrotoxicity. We present an explainable QSAR framework that predicts urinary tract toxicity of defined chemicals using mechanistically relevant 2D molecular descriptors. A curated set of 209 structurally diverse compounds was annotated by dose-based EPA LD₅₀ thresholds, ensuring toxicological relevance and chemical definition. Following rigorous preprocessing, 1444 descriptors were reduced to 11 variables capturing polarity, hydrogen-bonding capacity, and charge distribution. Twelve machine-learning algorithms were evaluated, and Random Forest performed best (accuracy = 84.13 %, AUC = 0.86). SHAP analyses provided global and compound-level explanations, implicating descriptor patterns consistent with epithelial barrier disruption, oxidative stress, and altered membrane permeability as drivers of toxicity. External test performance and applicability-domain assessment supported robust generalization across chemical space, with performance metrics comparable to those reported for earlier urinary and renal toxicity models. This work provides a transparent in silico tool for early safety evaluation that complements existing approaches by emphasizing mechanistic interpretability, supporting risk assessment for chemicals in the human environment, and fostering 3Rs by reducing reliance on animal testing. The models enable rapid screening and mechanistic interpretation to inform safer-by-design chemistry and regulatory decision-making.