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

Area of Science:

• Biochemistry
• Computational Biology
• Proteomics

Background:

• Protein glycosylation is a vital post-translational modification with significant roles in biological processes and disease.
• Glycoproteomics analysis using mass spectrometry faces challenges due to the structural complexity and heterogeneity of glycopeptides.
• Current glycopeptide identification tools often underutilize spectral intensity data, limiting their ability to differentiate between similar glycopeptides.

Purpose of the Study:

• To develop an innovative deep learning framework, GPMassSimulator, for accurate prediction of intact N-glycopeptide tandem mass spectrometry (MS/MS) spectra and retention time.
• To enhance the discrimination capabilities of glycopeptide identification tools by integrating peptide sequence and glycan structure information.
• To improve the sensitivity and accuracy of glycopeptide identification in complex biological samples.

Main Methods:

• Developed GPMassSimulator, a deep learning framework utilizing the GpepFormer module to represent and integrate peptide sequences and glycan structures.
• Employed a Prediction module within GPMassSimulator to generate theoretical MS/MS spectra and retention times for glycopeptides.
• Validated the model's performance on a benchmark dataset, including experiments distinguishing similar glycan compositions and isomeric structures.

Main Results:

• GPMassSimulator achieved 97.1% identification accuracy in distinguishing similar glycan compositions.
• The framework demonstrated more accurate Top-1 identifications for isomeric structures compared to existing approaches.
• A rescoring experiment on pGlyco3 data highlighted a significant improvement in the sensitivity of GPMassSimulator for glycopeptide identification.

Conclusions:

• GPMassSimulator offers a powerful deep learning approach for accurate prediction of N-glycopeptide MS/MS spectra and retention times.
• The framework effectively captures complex dependencies between peptide sequences and glycan structures, enhancing glycopeptide identification.
• GPMassSimulator shows significant promise for advancing glycoproteomics research and applications in disease state analysis.