박미나 (한국기초과학지원연구원)

박미나 (한국기초과학지원연구원)

김진영 (한국기초과학지원연구원)

황희연 (한국기초과학지원연구원)

Protein fucosylation is a key post-translational modification affecting protein structure, stability, and interactions. N-glycopeptide complexity arises from various combinations of HexNAc, Hex, Fuc, and Sia. Fucosylation is classified into core and outer types, both linked to cancer, immune responses, and protein regulation, requiring precise structural analysis. This study presents a method to classify N-glycopeptide fucosylation into none, core, outer, and dual types using deep neural networks (DNN) and support vector machines (SVM). To classify fucosylation types, we selected training and test sets from over 1,320 N-glycopeptide MS/MS spectra derived from immunoglobulin G (IgG) and alpha-1-acid glycoprotein (AGP). The N-glycopeptide MS/MS spectra were identified characteristic fragment ions of N-glycopeptides, and peak m/z and intensity values were applied to machine learning models. Various hyperparameters were tested to optimize performance, and the model was validated on human plasma samples to classify fucosylated N-glycopeptides. DNN and SVM approaches accurately predicted fucosylation types in complex plasma samples, demonstrating the effectiveness of combining machine learning with MS/MS analysis.