BACKGROUND & AIMS:The quality and market value of the medicinal herb Atractylodes lancea (AL) are critically dependent on its variety, geographical origin, and production mode. To combat adulteration and ensure efficacy, we developed a novel multi-platform analytical strategy integrated with machine learning to establish a robust traceability model for variety discrimination, geographical origin determination, and production mode identification of AL and identify the key chemical indicators responsible for its authentication.
RESULTS:Significant differences were found in trace element concentrations and isotopic ratios among samples. AL's main flavors were spicy, sweet, and fruity, with terpenoids as key aroma contributors. OPLS-DA identified key indicators for tracing AL's variety, including eleven trace elements (e.g., V, Al) and eight volatile compounds (e.g., β-Sesquiphellandrene, 2-Pinen-10-ol). For tracing AL origins, ten trace elements (e.g., Sr, Cr), two stable isotopes (δ13C, δ15N), five flavor components (e.g., 2-ethyl-3,6-dimethylpyrazine, 2-Pentadecanone), and twenty-six volatile components (e.g., γ-Gurjunene, β-Bisabolene) were identified. Furthermore, three trace elements (Mg, Li and Pb), two isotopes (δ13C and δ15N), two flavor components (α-Pinene and n-Nonylcyclohexane), and two volatile components (α-Copaene and α-Curcumene) were identified as key indicators for tracing AL's production modes. Finally, among the nine machine learning algorithms evaluated, LightGBM demonstrated superior performance, achieving a traceability accuracy of 95.28 ± 3.01%.
CONCLUSION:The multi-platform data fusion strategy presents a thorough and dependable approach to quality control for Atractylodes lancea. This method establishes a precise, efficient, and adaptable framework, demonstrating substantial potential for application to other high-value botanicals and complex natural products.