Origin-dependent chemical profiles and quality differentiation of candied Prunus mume: Insights from metabolomics, GC–MS volatile profiling, and interpretable machine learning

Abstract

Prunus mume is widely processed into candied products, but the food-chemical basis of origin-related quality variation remains unclear. Fruits from fourteen Chinese origins were processed into candied P. mume (CPM) under a unified candying regime and analyzed using quality-trait assays, gas chromatography–mass spectrometry (GC–MS) profiling of volatile organic compounds (VOCs), untargeted metabolomics, relative odor activity value (ROAV) analysis, weighted correlation network analysis (WCNA), and random forest–Shapley additive explanations (RF–SHAP). CPM showed origin-dependent differences in sugar–acid traits, antioxidant capacity, VOC profiles, and non-volatile metabolite profiles. ROAV and WCNA revealed shared aroma-active contributors and origin-associated volatile modules. The non-volatile metabolite profile showed abundance differences within a shared metabolite backbone, involving flavonoid glycosides, phenolic derivatives, carbohydrates/glycosides and organic-acid-related compounds. RF–SHAP prioritized differentiating metabolites, while cross-layer integration linked antioxidant-related variation to phenolic/flavonoid metabolism and sugar–acid variation to carbohydrates and organic acids. These findings provide a chemical basis for origin-oriented quality evaluation of CPM.