Chrysanthemum (Chrysanthemum morifolium Ramat.) has been recognized as both a food and medicinal substance in China since 2002 and possesses antioxidant, anti-inflammatory, antibacterial, and immunomodulatory activities. Previous studies suggest that Chrysanthemum may alleviate skin lesions resembling atopic dermatitis (AD); however, its underlying mechanisms remain unclear. In this study, we integrated network pharmacology and machine learning to systematically explore the potential mechanisms of Chrysanthemum in AD treatment. Four algorithms-Random Forest (RF), Lasso regression with cross-validation (LassoCV), Elastic Net (EN), and Extreme Gradient Boosting (XGB)-were compared, among which the XGB model achieved the best performance (accuracy = 0.9393). Further analysis identified 15 optimal features, two core targets (PTGS2 and MMP9), and one critical pathway (NF-κB signaling). To experimentally validate these findings, HaCaT keratinocytes were co-stimulated with TNF-α and IFN-γ to establish an in vitro inflammatory model, and co-treatment with three major flavonoids from Chrysanthemum-Acacetin, Diosmetin, and Chryseriol-significantly suppressed cytokine-induced COX-2 overexpression and reduced NF-κB p65 phosphorylation, confirming their inhibitory effects on NF-κB activation. These results were consistent with molecular docking and dynamics simulations, which demonstrated that these flavonoids, along with celecoxib, could stably bind to COX-2, thereby enhancing system stability and reducing residue fluctuations at the binding interface, revealing the molecular basis by which Chrysanthemum alleviates AD and supporting its modernization and therapeutic potential.
Keywords: Chrysanthemum; atopic dermatitis; machine learning; molecular dynamics simulation; network pharmacology.