Predictive q-RASAR modeling of intrinsic toxicity of pharmaceuticals and bioactive compounds in cats

Pharmaceuticals and bioactive compounds can be toxic when exposure exceeds safe thresholds. Domestic cats (Felis catus) share food, water, and indoor environments with humans and exhibit unique metabolic sensitivities, making them a relevant sentinel species for identifying hazardous chemical structures in shared human-animal microenvironments. Despite this relevance, systematic computational frameworks for feline toxicity prediction remain limited. In this study, novel quantitative structure-activity relationship (QSAR) and quantitative read-across structure-activity relationship (q-RASAR) models were developed to predict the Lowest Lethal Dose (LDLo) of diverse bioactive compounds using a curated feline dataset. These models identified influential molecular fragments (toxicophores and non-toxicophores) that govern chemical lethality, providing meaningful insights for safer drug design and environmental risk evaluation. To explore the non-linear structure-toxicity relationships, multiple modeling approaches were applied, including Random Forest (RF), Support Vector Machine (SVM), Neural Networks (NN), Gradient Boosting (GB), and Stochastic Gradient Descent (SGD). Comparative results demonstrated that the PLS-based q-RASAR model achieved the best predictivity (R² = 0.764, Q² = 0.727, Q²_F1 = 0.760, Q²_F2 = 0.760) with the lowest predictive error, confirming its robustness and accuracy. Further screening of the Pesticide Properties Database (PPDB) and DrugBank database effectively differentiated highly toxic and less toxic compounds. This integrative framework offers an early indication of hazardous chemical structures in a metabolically sensitive sentinel species, providing contextual insights useful for broader chemical safety assessment.