Machine learning based on metabolomics to discriminate Wuyi rock tea production areas and “rock flavor” substances

Zheng Peng, Wenmiao Wu, Chengjian Wu, Zhijun Zhao, Jian Chen, Juan Zhang,
Machine learning based on metabolomics to discriminate Wuyi rock tea production areas and “rock flavor” substances,
Food Chemistry: X,
Volume 31,
2025,
103194,
ISSN 2590-1575,
https://doi.org/10.1016/j.fochx.2025.103194.
(https://www.sciencedirect.com/science/article/pii/S2590157525010417)
Abstract: The “rock flavor” quality of Wuyi Rock Tea varies across production areas, but scientific classification criteria for production areas and a comprehensive understanding of the chemical basis of “rock flavor” remain limited. This study integrated metabolomics and machine learning to systematicallyanalyze the volatile metabolite profiles of 137 Wuyi Rock Teasamples (Zhengyan, Banyan, and Waishan productions) and established a high-precision random forest model (99 % accuracy) for production area discrimination. Feature importance analysis identified Zhengyan production markers as hotrienol, dihydroactinidiolide, benzyl alcohol, and trans-nerolidol.Banyan production markers as hotrienol, benzyl alcohol, trans-nerolidol, and heptanal,and Waishan production markers as methyl decanoate, (Z)-hept-4-enal, and 2,4-heptadienal. This study innovatively developed a volatile metabolite fingerprint-based system for Wuyi Rock Tea production area authentication and elucidated the key chemical foundations of “rock flavor,” providing theoretical support for geographical indication protection and processing optimization.
Keywords: Wuyi rock tea; Metabolomics; Machine learning; Production area discrimination; Rock flavor