Nontargeted metabolomics‐based discrimination of the geographical origin of P teridium aquilinum using liquid chromatography–high‐resolution mass spectrometry and machine learning

BACKGROUND: Distinguishing the geographical origins of agricultural products is essential for food safety, quality, and consumer protection. In this study, nontargeted metabolomics based on liquid chromatography-quadrupole time-of-flight mass spectrometry was leveraged to distinguish bracken (Pteridium aquilinum) originating from South Korea and China. RESULTS: A total of 109 samples were analyzed, and 333 features were identified. Based on the fold-change and t-test analyses, a volcano plot was constructed, revealing the differences among constituent metabolites. Moreover, principal component analyses allowed for initial separation, whereas partial least squares discriminant analyses (PLS-DA) and orthogonal PLS-DA yielded a clear and discriminative model demonstrating high predictive performance; fatty acids and phenolic derivatives, including 2-decarboxyphyllocactin and propylene glycol stearate, were identified as key contributors to the discrimination of geographical origin. Additionally, a machine learning model - random forest - was employed for fivefold cross-validation and demonstrated 100% accuracy, sensitivity, and specificity in suggesting that fatty acids and phenolic derivatives were the key discriminants. CONCLUSION: These findings highlight the robustness of combining metabolic profiling and machine learning as an approach for authenticating bracken origin, as well as the considerable potential in applying this approach for the quality control of agricultural products in food markets.