Machine learning and cheminformatics-based Identification of lichen-derived compounds targeting mutant PBP4

Penicillin-binding protein 4 (PBP4) is essential in imparting significant β-lactam antibiotics resistance in Staphylococcus aureus (S. aureus) and the mutation R200L in PBP4 is linked to β-lactam non-susceptibility in natural strains, complicating treatment options. Therefore, discovering novel therapeutics against the mutant PBP4 is crucial, and natural compounds from lichen have found relevance in this regard. The aim of our study was to identify novel inhibitors against the R200L mutation by applying machine learning (ML) approach. Predictive classification models were developed using six machine learning algorithms to categorize lichen-derived compounds as either active or inactive. The models were evaluated using ROC curves, confusion matrices, and relevant statistical parameters. Among these, the Extra Trees algorithm showed superior predictive accuracy at 81%. The model identified 115 potentially active compounds from lichen, which were further evaluated for drug-likeness and structural similarity to β-lactam antibiotics. The top 23 compounds, showing similarity to β-lactam drug, were subjected to molecular docking. Among the top 10 compounds, two compounds, Barbatolic acid and Orcinyl lecanorate, displayed promising results in 200 ns molecular dynamics (MD) simulations and MM-PBSA analysis, exhibiting better docking score compare to reference compound. Additionally, DFT calculations revealed negative binding energies and smaller HOMO-LUMO gaps for both compounds. The obtained results prove the utility of ML in screening natural compounds, and provide novel opportunities for the design of antimicrobial compounds in the future.