Integrated Metabolomics and Machine Learning Reveal Drying-Dependent Phenolic Markers in Asteraceae Edible Plants From Ulleungdo Island.

Ulleungdo Island, Korea, features a distinctive oceanic climate that supports a diverse array of edible wild plants traditionally processed through blanching and sun-drying. Despite their cultural and commercial significance, the effects of traditional drying methods on phytochemical composition remain inadequately understood. This study investigated processing-induced metabolic changes in three representative Asteraceae species-Aster pseudoglehnii, Cirsium nipponicum, and Solidago virgaurea-using integrated metabolomics and machine learning approaches. Untargeted UHPLC-HRMS/MS profiling identified 110 secondary metabolites across freeze-dried and sun-dried samples. Hierarchical clustering revealed that species identity was the primary determinant of chemotype, while processing induced consistent quantitative shifts within each species. Traditional processing (blanching followed by sun-drying) significantly reduced photo- and enzymatically labile flavonoid glycosides, whereas caffeoylquinic acid derivatives were relatively preserved. Random Forest analysis combined with SHAP interpretation identified caffeoylquinic acids, including chlorogenic acid and dicaffeoylquinic acid isomers, as processing-stable chemotaxonomic markers, while glycosylated flavonoids served as sensitive indicators of processing intensity. Targeted quantification confirmed these trends and revealed species-specific phenolic allocation patterns. Overall, this study demonstrates that integrating metabolomics with explainable machine learning enables robust identification of processing-dependent marker compounds and provides a biochemical foundation for traditional drying practices and marker-based quality control strategies for Ulleungdo wild vegetables.