Related Experiment Video
Updated: Aug 6, 2026
2D-HPLC-MS Technology Combined with Molecular Network for the Identification of Components in Tibetan Medicine Aconitum pendulum
Published on: December 8, 2023
Recognition Analysis of Ginseng Radix et Rhizoma and Confused Varieties according to Chemometrics and Machine
Tingting Cao1, Jiyu Zhao1,2, Xuweisheng Ji2
1School of Pharmacy, Jiangsu Food & Pharmaceutical Science College, Huai'an 223003, China.
Background:
Ginseng radix et rhizoma (GRER) is widely used in China and around the world. Due to the shortages, high prices, and profitability, the market is also full of its confused varieties. Notoginseng radix et rhizoma, panacis quinquefolii radix, ginseng folium, and ginseng radix et rhizoma rubra are GRER's common confused varieties, which affected the market order and drug safety in the case of incorrect use. Therefore, it is very important to realize the recognition analysis of GRER and its confused varieties.
Methods:
GRER and its confused varieties were studied using ultraperformance liquid chromatography-quadrupole time-of-flight mass spectrometry to convert the data matrix. Then, the data matrix was used to conduct principal component analysis (PCA) and partial least-squares discriminant analysis (PLS-DA). At the same time, the data matrix was also used to construct data identification models based on machine learning, and the best model was screened for external appraisal and verification analysis. Moreover, the differential chemical components were analyzed on the basis of feature screening.
Results:
the data matrices contained 1971 chemical components and PCA and PLS-DA results showed that GRER cannot be clearly distinguished from confused varieties. However, all the identification models based on K-nearest neighbor, artificial neural network, support vector machine, naive bayes, and random forest (RF) had an excellent identification effect with area under the curve (AUC) ≥0.980, accuracy ≥0.800, and precision ≥0.870 in which the RF model showed the best recognition effect with AUC = 1.000, accuracy = 1.000, and precision = 1.000. Twenty batches of test samples were accurately identified by external verification, and the correct rate was 100%. Pseudoginsenoside F11, ginsenoside Rb2, ginsenoside Ro, and so on were the important distinguishing chemical markers.
Conclusion:
compared with chemometrics, machine learning presented better identification results. Moreover, the RF model has the best recognition effect, which helps to identify the GRER and its confused varieties. In addition, the characteristic components based on feature screening also contributed to the discrimination of the GRER and its confused varieties.
