Integrated Machine Learning and Structure-Based Virtual Screening Identifies Natural Product Targeting 50S Ribosome Inhibitory Activity Against Cutibacterium acnes

Acne vulgaris is a prevalent inflammatory disease of the pilosebaceous unit in which <i>Cutibacterium acnes</i> (<i>C. acnes</i>) contributes to lesion initiation and persistence, supporting antibacterial interventions as a component of clinical management. Given the essential role of the 50S large ribosomal subunit-particularly 23S rRNA sites in the peptidyl transferase center and nascent peptide exit tunnel-in <i>C. acnes</i> protein synthesis and viability, targeting the 50S offers an effective path to lead discovery for acne treatment. Here, we present an integrated computational-experimental workflow to identify anti-<i>C. acnes</i> candidates from a 186,659-compound natural product library. Curated 50S/23S ligands trained and validated two ML-QSAR regression models built on different molecular fingerprints (MACCS keys and PubChem 2D) to predict anti-<i>C. acnes</i> activity and rapidly triage the library. Compounds were further screened by ADMET filtering and structure-based docking to 23S rRNA pockets, followed by cluster and interaction analysis. Among six experimental hits, three compounds exhibited MICs against <i>C. acnes</i> of ≤8 μg/mL, with tripterin, a pentacyclic triterpenoid, being the most potent (0.5-2 μg/mL across two acne-relevant strains). Collectively, these results indicate that a 50S ribosomal-focused, multistage computational screening workflow, integrated with in vitro assays, efficiently prioritizes compounds with quantifiable anti-<i>C. acnes</i> activity across a broad range of natural products.