Screening the grading markers and their application in the grade discrimination of Gastrodiae Rhizoma using metabolomics and machine learning

• 101 primary and secondary metabolites in GR samples were quantified • 25 featured metabolites related to the GR grades were selected • Parishins were the main metabolites influenced the quality and activity of GR • Six machine learning models were established to quickly identify the GR grades • SVM-L, SVM-P, and SVM-R showed best performance in GR grades prediction Gastrodiae Rhizoma (GR), the tuber of the Gastrodia elata Blume, is a food with functional properties. The grades of GR are closely related to its biological activity, and are mainly identified by their appearance characteristics, mainly individual weight. At present, few studies focus on the content difference of chemical components in different grades of GR. In this study, different batches of GR samples were divided into four grades based on their weight. Subsequently, a total of 101 metabolites were identified using UPLC-HRMS, and 25 grading markers were screened, such as parishin A, parishin B, parishin F, etc. Meanwhile, antioxidant activity of GR samples was evaluated, and parishin A and parishin B were closely related to their bioactivities. Finally, six machine learning models were established to identify the GR grades, and support vector machine (SVM) linear kernel, SVM polynomial kernel, and SVM radial kernel showed best performance index. This study revealed the intrinsic reason for the discrimination of GR samples based on their individual weight, provided a new strategy to improve and control the quality of GR samples, and offered potential clues for assessing the counterfeiting GR samples with high grades in the market.