Authenticity discrimination of Pulsatillae Radix based on dry-process REIMS fingerprinting combined with machine learning / 中国中药杂志

In this study, rapid evaporative ionization mass spectrometry(REIMS) fingerprints of 388 samples of roots of Pulsatilla chinensis(PC) and its common counterfeits, roots of P. cernua and roots of Anemone tomentosa were analyzed based on REIMS combined with machine learning. The samples were determined by REIMS through dry burning, and the REIMS data underwent cluster analysis, similarity analysis(SA), and principal component analysis(PCA). After dimensionality reduction by PCA, the data were analyzed by similarity analysis and self-organizating map(SOM), followed by modeling. The results indicated that the REIMS fingerprints of the samples showed the characteristics of variety differences and the SOM model could accurately distinguish PC, P. cernua, and A. tomentosa. REIMS combined with machine learning algorithm has a broad application prospect in the field of traditional Chinese medicine.

Drug analysis based on scalable moving-window similarity and Bayesian method by Raman spectroscopy / 药学实践杂志

Objective To propose scalable moving-window similarity combined with Bayesian for rapid discriminating low active pharmaceutical ingredient(API)signal drugs(LAPIDs).Methods The scalable moving-window similarity method was employed by setting the window size dynamically according to API′s peak width.In each window,the correlation coefficient (CC)of API′s peak spectrum signal with LAPID′s spectrum and LAPID′s spectrum with excipient′s spectrum were calculated respectively.The LAPIDs discrimination model was established by choosing windows with most contribution of the API spec-tral signal to the LAPID spectrum as variables for Bayesian discriminant model.Results The accuracy rate of LAPIDs discrim-ination model for discriminating LAPIDs was 94.7%.The accuracy rate of the model for discriminating testing samples was 95.6%.Conclusion Bayesian discrimination model based on scalable moving-window similarity and Bayesian algorithm can quickly discriminate LAPIDs.