Machine Learning Classification Models for the Design of Novel Nitroimidazole Derivatives with Anti-Trichomonas vaginalis Activity.

Trichomonas vaginalis is the pathogen responsible for trichomoniasis, a widespread sexually transmitted infection worldwide. Treatment currently relies solely on nitroimidazole-class drugs, such as metronidazole and tinidazole, to which resistance is increasingly reported. When resistance or treatment failure occurs, higher doses are prescribed, which in turn heightens the risk of toxic side effects. Therefore, discovering new therapeutic options is urgently needed. To address this, quantitative structure-activity relationship (QSAR) models were developed using nitroimidazole derivatives to analyze the relationship between various chemical structures and their efficacy against the parasite, selectivity between two different strains, and potential cytotoxicity against HeLa cells. Three distinct models were selected, demonstrating strong performance in both internal and external validation (external MCC values ranging from 0.583 to 0.849). These classification models were then applied to predict the biological activity of 57 nitroimidazole compounds, aiming to identify promising new candidates for treating T. vaginalis infections. Three selected hit compounds were synthesized and experimentally validated against T. vaginalis as well as HeLa cells. All compounds were active, with IC50 values ranging from 0.17 to 48.92 μM, with no cytotoxicity for HeLa cells observed at the tested concentrations, showing them to be promising candidates for preclinical evaluation.