Mixed-mode multidimensional polysaccharide fingerprint coupled with hierarchical data fusion for the precise authentication of Gastrodia elata Blume subtypes.

Gastrodia elata Blume is a vital herb requiring precise subtype discrimination for market authentication and traceability. However, the current analytical focus on small molecules limits the extraction of highly discriminative polysaccharide features, making chemometric authentication under class-imbalanced conditions highly challenging. To address these limitations, we established a robust multi-detector size-exclusion chromatography platform utilizing a 5 mM NH₄Ac mobile phase to induce a mixed-mode separation, effectively resolving the cryptic structural heterogeneities and multidimensional properties of the bioactive polysaccharides. Furthermore, we systematically evaluated nine machine learning algorithms, comparing hierarchical versus direct architectures to mitigate class imbalance. For multimodal data fusion, we assessed chromatographic profile, physicochemical expert, and fused feature strategies, utilizing PCA, RFECV, and SFM for dimensionality reduction. Results demonstrated that the hierarchical architecture successfully resolved class imbalance. The KNN model utilizing PCA-reduced fused features achieved optimal performance (Accuracy: 0.8927, F1-Macro: 0.8515). More importantly, interpretable analysis using SHapley Additive exPlanations and PCA loadings revealed a robust chemical decision mechanism: the model prioritizes macroscopic molecular weight (PC1) to identify Gastrodia elata Bl. f. glauca S. Chow (WTM), while utilizing microscopic structural details (PC3/PC5) to distinguish Gastrodia elata Bl. f. elata (HTM) and their hybrids (WHTM). This interpretable chemometric strategy enables precise Gastrodia elata subtype authentication and offers a reliable reference for analyzing polysaccharide-rich herbs.