Integration of machine learning-predicted collision cross sections for the discrimination of flavonol 3-O- and 7-O-glycoside isomers.

Flavonols are prominent bioactive metabolites whose efficacy depends on glycosylation patterns, necessitating precise isomeric characterization. However, unambiguous identification of flavonol glycosides remains challenging due to the limited availability of authentic standards and the low intensity of fragment ions. Here, a high-accuracy machine learning model was developed for collision cross section (CCS) prediction, focusing on the resolution of flavonol 3- and 7-O-glycoside isomers. A support vector machine model was trained using a curated flavonoid dataset and Chemistry Development Kit molecular descriptors, achieving excellent performance, with 95% of the external test set predicted within a 3% error margin. The model was further validated through the characterization of flavonol glycosides in tea. Even at trace levels where MS/MS fragmentation was insufficient, the synergistic integration of predicted CCS provided the orthogonal evidence for isomeric discrimination. This study underscores the potential of machine learning-derived CCS for providing improved confidence in the annotation of flavonol glycosides.