GPMassSimulator: A Graphormer-Based Method for Glycopeptide MS/MS Spectra Prediction

Protein glycosylation is a critical post-translational modification involved in numerous biological processes and disease states. While mass spectrometry has emerged as the primary tool for glycoproteomics analysis, the structural complexity and heterogeneity of glycopeptides pose significant analytical challenges. Existing glycopeptide identification tools primarily rely on mass matching, underutilizing intensity information from mass spectra, which limits their ability to discriminate between similar glycopeptides (glycopeptides bearing analogous glycans/peptide backbones). Here we present GPMassSimulator, an innovative deep learning framework for accurate prediction of intact N-glycopeptide tandem mass spectrometry (MS/MS) spectra and retention time. GPMassSimulator employs the GpepFormer module to effectively represent and integrate both peptide sequences and glycan structures, capturing their complex dependencies. The integrated representation is then passed through the Prediction module to generate the theoretical MS/MS spectra and the retention time of the glycopeptides. Our method demonstrated an outstanding performance on the benchmark set. In the experiment distinguishing similar glycan compositions, GPMassSimulator achieved an identification accuracy of 97.1%. Furthermore, in distinguishing isomeric structures, our method achieved more accurate Top-1 identifications than the current approaches. Additionally, the rescoring experiment on pGlyco3 highlighted the significant improvement in the sensitivity of our model for glycopeptide identification. These excellent results showcased the promising potential of our approach in glycoproteomics.