Machine learning-enabled metabolomics for geographical authentication of Lonicera japonica via UHPLC-Q-TOF-MS/MS and SHAP interpretation.

Geographical traceability is vital for ensuring the authenticity and quality of food and medicinal plants. Lonicera japonica Thunb. (Jinyinhua, JYH) is widely consumed for its pharmacological and nutritional benefits, yet its quality is strongly origin-dependent and prone to adulteration. Here, we integrated ultra-high-performance liquid chromatography-quadrupole-time-of-flight tandem mass spectrometry (UHPLC-Q-TOF-MS/MS)-based metabolomics with machine learning for precise authentication of JYH from eight production areas. Metabolomic profiling with molecular networking characterized 156 analytical samples, annotating 136 metabolites. Among 13 models tested, CatBoost achieved superior performance (AUC = 0.99). Shapley Additive Explanations (SHAP) analysis enabled global and local interpretation of classification features. A refined multi-marker screening identified seven key discriminative marker compounds that retained high accuracy (0.87, AUC = 0.99) and showed clear origin-related abundance patterns. This transferable and interpretable workflow offers a robust solution for JYH traceability and quality control, and provides a paradigm applicable to authenticity verification of other medicine-food homologs.