Unraveling the variations and relationships between phytochemical constituents and anti-inflammatory potentials of three Zanthoxylum species through metabolomics and explainable machine learning

Zanthoxylum plants are usually deciduous trees or shrubs of Rutaceae, as industrial crops widely cultivated in China. Three representative Zanthoxylum species, Z. bungeanum, Z. schinifolium, and Z. armatum, are well-known for their pericarps rich in bioactive compounds and can be utilized as functional additives or phytomedicine. These mentioned Zanthoxylum species are distributed and intermixed with each other as "Zanthoxylum" on markets, which creates obstacles for consumers to identify them. In the present study, 49 batches of Zanthoxylum extracts were examined for anti-inflammatory activity and then employed in acquiring corresponding multi-constituent profiles by UHPLC-Q-Orbitrap-HRMS and UHPLC-MS/MS. The obtained data were further used for the construction of extreme gradient boosting (XGBoost) classification and regression models. A Shapley additive explanations (SHAP) algorithm was subsequently applied to interpret the XGBoost classification/regression model output for desirable information. Isoquercitrin, rutin, and arctigenin were identified as the important chemical markers for discriminating Zanthoxylum origins through the classification algorithm. Furthermore, six constituents, consisting of hydroxy-γ-sanshool, avicularin, bergapten, rutin, scopoletin, and narcissoside, were screened for their major anti-inflammatory activity by modeling using the regression approach. Unraveling the diversities and connections of constituents and biological activities in these three species could contribute to their reasonable industrial and medicinal applications.