Machine learning modeling for optimizing fermentation process of Dendrobium officinale with improved antioxidant property and metabolic changes.

Lactic acid bacteria (LAB) fermentation may affect chemical accessibility and can develop antioxidant properties of Dendrobium officinale (DO). This study integrated nine machine learning (ML) algorithms with Response Surface Methodology (RSM) to optimize polyphenol enrichment. XGBoost emerged as the superior predictive model (R2 = 0.982 ± 0.006, RMSE = 2.92 ± 0.076, and MAE = 2.35 ± 0.062), significantly outperforming the quadratic RSM model. LAB fermentation effectively enhanced bioactive small molecule production and radical scavenging activity (DPPH and ABTS) without reducing cell viability under the tested in vitro conditions. Untargeted metabolomics identified profound metabolic remodeling, characterized by the upregulation of 31 flavonoids (notably 3-hydroxyflavone) and enrichment of bioactive phenolic acids like 1,3-dicaffeoylquinic acid. Importantly, fermentation significantly remodeled alkaloid metabolism, including spartein-5-ol and ambelline. These results demonstrate that ML-guided fermentation is a promising strategy for the functional valorization of medicinal and edible homology resources and provides a data-driven framework for fermentation process optimization.