Untargeted UHPLC–Q-Orbitrap-MS/MS profiling of six bamboo taxa for comparative chemical discrimination and SHAP-guided prioritization of antioxidant-associated metabolites

Bamboo leaves are phenolic-rich botanical materials with potential food and health applications, but their inter-taxon chemical heterogeneity complicates chemotaxonomic differentiation and chemistry-based quality evaluation. Here, untargeted ultra-high-performance liquid chromatography coupled (UHPLC)–Q-Orbitrap–MS/MS profiling combined with explainable machine learning was applied to 60 batches of bamboo leaves from six taxa to characterize chemical differences and identify metabolites associated with antioxidant assay variation. Fifty-eight recurrent metabolites were annotated, mainly C-glycosyl flavones, tricin derivatives, and phenolic acids. Supervised models outperformed principal component analysis (PCA) for taxon discrimination, with random forest (RF) achieving the best five-fold cross-validated performance (accuracy 0.950, area under the curve (AUC) 0.998) and yielding, together with support vector classification (SVC), a conservative 10-metabolite marker panel. Antioxidant responses were evaluated by 2,2-diphenyl-1-picrylhydrazyl (DPPH), 2,2′-azino-bis (3-ethylbenzothiazoline-6-sulfonic acid) (ABTS), and ferric reducing antioxidant power (FRAP) assays, and RF showed the best internal prediction (R 2 0.950–0.973). SHapley additive exPlanations (SHAP) analysis linked antioxidant variation mainly to a chemically coherent subset of flavonoids and hydroxycinnamate derivatives. Overall, this study provides a chemistry-based workflow for bamboo leaf taxon discrimination and antioxidant marker prioritization.