Machine-Learning Approach to Increase the Potency and Overcome the Hemolytic Toxicity of Gramicidin S

Antibiotic resistance is a global health crisis, with multidrug-resistant pathogens like methicillin-resistant Staphylococcus aureus (MRSA) demanding next-generation therapeutics. Tackling this silent pandemic requires innovative strategies beyond traditional drug discovery. We present a machine-learning (ML)-driven computational pipeline for redesigning FDA-approved drugs, applied here to the cyclic antibiotic gramicidin S, historically limited to topical use due to hemolytic toxicity. Leveraging a proprietary analogue data set, the model identified key molecular descriptors linked to potency and safety, yielding several potent, nontoxic candidates. Peptide 2 expanded the therapeutic window 42-fold, eliminating hemolysis at bactericidal doses. Peptide 9 achieved a significant 2-fold increase in potency against MRSA (MIC: 2 μg/mL) and improved the therapeutic index 6-fold. These analogues represent the most significant enhancement to the safety and efficacy of gramicidin S to date, enabling potential systemic MRSA treatment. Our ML-guided framework offers a powerful, generalizable platform for optimizing other FDA-approved drugs across therapeutic areas.