Wine processing as a chemical driver for reprogramming the metabolomic profile and color phenotype of Radix Paeoniae Rubra: an integrative molecular networking and thermochemical study.

Wine processing is a long-established strategy in herbal food manufacturing, yet the molecular rules by which rice wine reshapes complex herbal matrices and translates chemical remodeling into visible quality traits remain poorly defined. Here, Radix Paeoniae Rubra (Rpr), a representative medicinal-edible herb, was used to construct an integrated "processing-metabolism-phenotype-mechanism" framework by combining a four-factor, multi-level processing design with color phenotyping, UPLC-Q-TOF-MS-based metabolomics, multivariate statistics, machine-learning-assisted feature selection, feature-based molecular networking, RMG thermochemical estimation, and IBM RXN-assisted reaction prediction. Wine processing triggered extensive reprogramming of the Rpr metabolome, manifested by enhanced early-eluting polar signals, increased chromatographic complexity, and coordinated shifts in CIE L*a*b* color parameters and visible reflectance. Through a multidimensional consensus strategy integrating PLS-DA, random forest, SVM-ROC, and FDR-corrected ANOVA, 399 processing-responsive features were retained, including 326 positive-ion and 73 negative-ion features. Factor-wise PERMANOVA/RDA and exploratory trajectory modeling identified stir-frying temperature as the factor most strongly associated with metabolic divergence, with a model-estimated transition region around 90 °C rather than a precisely resolved critical temperature. Feature-based molecular networking suggested a hierarchical transformation cascade involving lipid/ester hydrolysis, glycoside degradation, oxidation, and Maillard-like nitrogen incorporation. Rice wine was interpreted as a potential exogenous chemical input supplying organic acids, amino acids, ethanol-derived acetaldehyde, and aromatic small molecules. Nitrogen-containing features 9391 and 2165 were assigned provisional structures and used to formulate testable transformation hypotheses. RMG-estimated thermochemical descriptors and IBM RXN predictions were treated only as auxiliary computational evidence compatible with, but not experimental validation of, the proposed pathways. Overall, this study demonstrates that wine processing of Rpr represents an exogenous-endogenous coupled chemical reprogramming process driven by concurrent degradation and molecular reconstruction. By linking processing parameters, color phenotype, metabolomic remodeling, and reaction mechanisms, this work provides an interpretable AI/ML-assisted framework for mechanistic decoding, quality evaluation, and process optimization in herbal food manufacturing.