Optimization of flavonoids extraction and elucidation of antioxidant mechanisms in Dendrobium flexicaule using metabolomics and machine learning

Recent studies have demonstrated that flavonoids constitute a major class of bioactive compounds in Dendrobium species, contributing significantly to their pharmacological properties. However, the underutilization of flavonoids from Dendrobium is largely attributable to two interrelated bottlenecks: (1) the absence of systematic phytochemical screening to identify high-flavonoid germplasm resources, and (2) the lack of robust, scalable extraction protocols optimized for both yield and reproducibility. To address these limitations, this study first employed untargeted metabolomics to comparatively characterize the flavonoid profiles across four representative Dendrobium species. Subsequently, we developed an integrated optimization framework combining single-factor experimental screening, response surface methodology (RSM), and machine learning–based predictive modeling to rationally design and validate an efficient, high-yield flavonoid extraction protocol. Results revealed that Dendrobium flexicaule exhibited the highest total flavonoid content among the four investigated species. Under the optimized extraction conditions, 94 % (v/v) ethanol, 68 min extraction time, a material-to-liquid ratio of 1:50 (w/v), and 72 °C, the flavonoid yield reached 8.90 ± 0.17 mg/g dry weight. Among the machine learning models evaluated, the support vector regression (SVR) model demonstrated the strongest predictive accuracy, achieving an R 2 of 0.9893, an RMSE of 0.082 mg/g, and an MAE of 0.058 mg/g. However, untargeted metabolomic profiling of the optimized extract identified 34 flavonoids, with rutin as the most abundant compound, followed by eriodictyol and naringenin chalcone, both structurally confirmed by reference standards and spectral data. In vitro functional assays demonstrated that the extract exhibited dose-dependent antioxidant activity and significantly alleviated t-BHP induced oxidative stress in HepG2, mechanistically through activation of the Nrf2/ARE signaling pathway. Collectively, this study identifies Dendrobium flexicaule as a high-potential, flavonoid-rich botanical resource and establishes a robust, integrated framework combining untargeted metabolomics with machine learning–driven optimization to enhance both the efficiency of flavonoid extraction and the rigor of downstream bioactivity assessment.