Abstract
Targeting autoinducer-2 (AI-2) quorum sensing (QS) systems with dietary compounds represents a promising strategy to combat pathogens, yet mechanisms remain elusive. Here, we develop a machine learning-driven framework combining computational screening with multi-level experimental analysis to identify AI-2 quorum sensing interference molecules (QSIMs). A graph neural network (GNN)-based classifier (AI2IMpred) screens over 7000 phytochemicals and identifies dietary polyphenols as potent QSIMs. Using Salmonella Typhimurium LT2 as a model, we demonstrate that 17 polyphenols, including pterostilbene, 4-methylcatechol, pyrogallol, and 3-methylcatechol, directly target LuxS and LsrB, as confirmed by SPR assays, while molecular docking predicts potential interactions with TqsA and LsrR. Structural analysis revealed that para-substituted hydrocarbyl groups enhance LsrB binding. Furthermore, polyphenol-mediated regulatory effects have been validated via gene expression modulation, bacterial phenotypes, NCM460 cell adhesion/invasion, and cross-pathogen analysis (Escherichia coli, Staphylococcus aureus, and Pseudomonas aeruginosa). This work provides insights for the potential development of diet-based interventions using polyphenol compounds against drug-resistant pathogenic infections.