Detection of Artemisia mongolica floss adulteration in moxa floss: A strategy based on UPLC-Q/Orbitrap HRMS, chromatographic analysis, and machine learning.

Detecting adulteration in closely related herbal medicines is a challenge for quality control and market standardization. This study aimed to identify whether Artemisia mongolica floss was adulterated into moxa floss. First, ultra-performance liquid chromatography coupled with quadrupole Orbitrap high-resolution mass spectrometry was employed to analyze the chemical compositions of moxa floss and Artemisia mongolica floss. Combined with partial least squares discriminant analysis, four key differential biomarkers were identified. Subsequently, high-performance liquid chromatography (HPLC) methods were established for these biomarkers, providing the data foundation for subsequent adulteration discrimination model development. Moreover, four machine learning classification models were constructed: logistic regression, decision tree, random forest, and extreme gradient boosting to compare their ability to identify adulterated samples. The results indicated that the decision tree model exhibited strong generalization capability and stability. Based on the established HPLC method and decision tree model, the ratio range of peak areas of the biomarker 5-hydroxy-3',4',6,7-tetramethoxyflavone to eupatilin or jaceosidin can be used as an indicator to effectively identify adulterated samples. In summary, the study established an analytical strategy for identifying specific markers, thereby facilitating practical discrimination in cases of moxa floss adulteration.