High-accuracy glycan de novo prediction for N- and O-linked glycopeptides across multiple fragmentation techniques

N- and O-glycosylation are structurally diverse post-translational modifications that affect a range of biological functions. Glycoproteomics faces substantial challenges, particularly in the analysis of O-glycans due to their diversity compared to that of N-glycans. In addition, tandem mass spectrometry data patterns exhibit variability between different fragmentation methods. Existing de novo algorithms often lack sensitivity and are limited to sceHCD fragmentation, restricting their practical application. To address these limitations, we introduce DeepGlycan, a deep-learning-based method for de novo glycopeptide sequencing that captures relationships between glycopeptide spectra and fragment ions from both N- and O-glycans. DeepGlycan achieves over 92% glycan recall and around 95% glycan precision on N-glycopeptide spectra generated using both sceHCD and EThcD. In addition, it enables O-glycan de novo sequencing without additional training. Beyond benchmarking, DeepGlycan identifies an O-glycopeptide in mouse heart tissue whose assignment is supported by exoglycosidase treatment and comparison with a synthetic standard. Protein glycosylation is difficult to analyse across glycan classes and fragmentation methods. Here, the authors present DeepGlycan, a model that predicts N- and Oglycan compositions from tandem mass spectra and supports analyses across tissues and disease datasets.