A streamlined analytical strategy combining plant metabolomics and interpretable machine learning for detecting homologous adulteration in Panax notoginseng powder.

Panax notoginseng powder (PNP) is a high-value traditional Chinese medicine, yet persistent supply shortages have increased the risk of economically motivated adulteration. The covert homologous adulteration of PNP with its non-medicinal fibrous roots (FR) presents a formidable analytical bottleneck due to highly overlapping chemical profiles, leaving a critical gap in quality regulation. This study aimed to develop a highly accurate, interpretable, and deployable analytical strategy for homologous adulterant detection. Herein, we propose an integrated analytical approach combining LC-MS-based untargeted and targeted metabolomics, explainable machine learning (ML), and SHAP-guided feature selection to overcome the intractable challenge of authenticating PNP against homologous FR adulteration. Among the five ML models, XGBoost achieved the highest accuracy, with 100% accuracy for adulteration types. The SHAP analysis identified five quality markers (Q-markers), and a novel XGBoost model achieved 100% accuracy in classifying the Adulterated PNP category. Furthermore, notoginsenoside Fd (N-Fd) was biologically validated as an ecologically stress-driven defense biomarker uniquely enriched in FR. Application of the model to 84 commercial PNP and Hongyaopian samples identified a subset of products with FR-like ginsenoside profiles, highlighting its potential utility as a preliminary risk-screening tool for market surveillance. By conquering this ultimate homologous adulteration challenge, this computationally lightweight and interpretable paradigm establishes a cost-effective, modernized standard for routine industrial botanical quality control.