SAR Modeling to Predict Ames Mutagenicity Across Different Salmonella typhimurium Strains

Background: The Ames test, a biological assay employing various strains of Salmonella typhimurium, serves as a cornerstone in genetic toxicology for evaluating the mutagenic and potentially carcinogenic properties of chemical compounds. However, experimental testing is resource-intensive and time-consuming for screening the vast chemical space of existing and novel drug candidates in pharmaceutical development. Methods: To address this limitation, we have developed the Ames Mutagenicity Predictor web application, which predicts mutagenic activity in the Ames test for given structural formulas across a comprehensive panel of different bacterial strains. The application utilizes advanced structure-activity relationship (SAR) models generated by PASS (Prediction of Activity Spectra for Substances) v2024 software. The training set comprised 3250 compounds with experimentally determined mutagenicity across 69 different strains, compiled from peer-reviewed literature and established databases, and 4285 non-mutagenic compounds from the WWAD as negative examples. Results: Leave-one-out cross-validation (LOOCV) of the 69 strain-specific models yielded an average Invariant Accuracy of Prediction (IAP) of about 0.944, and for the unspecified mutagenicity, a value of 0.962 was obtained. Conclusions: These validated models have been integrated into a freely accessible web application Ames Mutagenicity Predictor that enables users to input compound structures through multiple formats: a built-in chemical editor, SMILES notation, or compound name search. The application generates comprehensive reports detailing the predicted probability of positive Ames test results for each individual strain, providing researchers with detailed mutagenicity profiles.