High-quality MS/MS spectrum prediction for data-dependent and data-independent acquisition data analysis.

Area of Science:

• Proteomics
• Mass Spectrometry
• Computational Biology

Background:

• Accurate prediction of peptide fragmentation spectra is crucial for interpreting mass-spectrometry-based proteomics data.
• Current methods struggle to accurately estimate fragment ion intensities due to incomplete understanding of ion generation.

Purpose of the Study:

• To develop a machine learning model for accurate prediction of peptide fragmentation patterns.
• To investigate the underlying factors influencing peptide fragmentation, such as sequence interactions.
• To demonstrate the utility of predicted spectra in analyzing proteomics datasets.

Main Methods:

• Utilized machine learning algorithms to predict peptide fragmentation patterns.
• Analyzed model outputs to identify key features driving fragmentation.
• Applied the developed models to both data-dependent and data-independent acquisition datasets.

Main Results:

• Achieved prediction accuracy within the uncertainty of experimental measurements.
• Discovered that peptide fragmentation is influenced by long-range interactions within the peptide sequence.
• Observed improved peptide identification rates in data-dependent acquisition and near-equivalence to experimental libraries in data-independent acquisition.

Conclusions:

• Machine learning provides a powerful tool for accurately predicting peptide fragmentation spectra.
• Understanding long-range interactions enhances the predictive power of fragmentation models.
• Predicted tandem mass spectrometry spectra can effectively substitute experimental libraries in data analysis.