Machine learning sensing coupled with critical components elucidates Peucedanum praeruptorum Dunn quality under multi-factors

Abstract

Peucedanum praeruptorum Dunn as a traditional Chinese medicine (TCM) with many clinical applications. Praeruptorin A (PA) and Praeruptorin B (PB) are quality markers. Recently, PB has attracted attention for its difficulty in satisfying Chinese Pharmacopoeia in Peucedanum praeruptorum Dunn. However, the association between the contents of PA and PB, as well as the factors influencing this association, remains unclear. Hence, we conducted a study on PA and PB of 538 batches of samples from the main production areas of Chinese provinces. A negative correlation between PA and PB was observed for the first time, with a wide range of fluctuation coefficients and poor quality stability. This relates closely to the main production areas and growth patterns. Particularly, the sum of PA and PB reveals the quality stability and maintains the satisfactory rate to avoid quality evaluation bias. Six machine learning algorithms were used to build the model after optimisation and evaluation. We found that the prediction accuracies for the DaoDi producing regions Anhui and Zhejiang reached 93.3 % and 87 % with Stacking and SVM, respectively. The kNN predicts wild and domestic species patterns with an accuracy of 93.3 %. Compared with traditional chemometrics, machine learning has absolute advantages. This study comprehensively revealed the quality formation factors of Peucedanum praeruptorum Dunn. It provides a scientific basis for improving the quality standard, grade evaluation and scientific supervision of Peucedanum praeruptorum Dunn.