Machine learning-guided discovery of covalent sortase A inhibitors targeting MRSA virulence

The global rise of methicillin-resistant Staphylococcus aureus (MRSA) has highlighted the urgent need for alternative therapeutic strategies beyond conventional bactericidal antibiotics. Targeting bacterial virulence rather than viability represents a promising approach to mitigate selective pressure and delay resistance development. Sortase A (SrtA), a membrane-associated transpeptidase responsible for anchoring virulence-associated surface proteins, is an attractive anti-virulence target due to its non-essential role in bacterial survival. Here, we report a machine learning-guided strategy for the discovery of novel covalent SrtA inhibitors based on a 1,2-benzoselenazol-3-one (BSEA) scaffold featuring a tunable electrophilic Se-N bond. A scaffold-aware classification model with a Tanimoto similarity constraint trained on 529 SrtA inhibitors enabled prospective virtual screening of over 35,000 BSEA and BTA derivatives, leading to a high hit rate of 89% upon experimental validation. Representative compounds exhibited submicromolar SrtA inhibition (IC₅₀ = 0.84-1.04 μM) while showing minimal effects on bacterial growth (MIC = 8-32 μM), indicating effective functional decoupling of virulence and viability. Mechanistic studies demonstrated time-dependent irreversible inhibition kinetics, supported by jump dilution assays and Nano-LC-MS/MS identification of covalent modification at the catalytic residue Cys184. These inhibitors effectively disrupted MRSA biofilm formation at sub-inhibitory concentrations and significantly improved host survival in a Galleria mellonella infection model. Collectively, this study establishes a data-driven framework integrating machine learning and covalent chemistry for anti-virulence drug discovery and provides promising lead compounds targeting SrtA to combat MRSA infections.