Metabolomic signatures and machine learning-based geographical origin discrimination of fresh tea leaves from Enshi Prefecture

Abstract

Geographical environment is an important source of variation in tea quality, yet the extent to which regional micro-environmental differences are reflected in the metabolic fingerprints of fresh leaves remains unclear. This study characterized 47 samples across six tea-producing counties in Enshi Prefecture. The results revealed distinct regional chemical differentiation, notably high amino acid levels in Hefeng and elevated polyphenols in Xianfeng (P < 0.05). Weighted Gene Co-expression Network Analysis (WGCNA) identified region-specific metabolic modules, highlighting enriched lipids and organic acids in Badong. Furthermore, after parameter optimization using five-fold cross-validation within the training subset, the Random Forest model achieved 100% accuracy in the internal test subset (AUC = 1.00), and SHAP analysis identified 20 candidate discriminant metabolites, including adenosine and geranic acid. These findings provide new insights into regional metabolic differentiation and highlight the potential of metabolomic fingerprints for geographical-origin discrimination of fresh tea leaves from Enshi Prefecture.