Integration of alternative fragmentation techniques into standard LC-MS workflows using a single deep learning model...

Area of Science:

• Mass Spectrometry
• Proteomics
• Computational Biology

Background:

• Collision-induced dissociation (CID) is the standard for bottom-up proteomics but struggles with complex samples.
• Limitations exist in characterizing post-translational modifications and proteoforms using CID.

Purpose of the Study:

• To develop an integrated mass spectrometry platform for automated multi-fragmentation techniques.
• To train a unified deep learning model for spectral prediction across fragmentation methods.
• To enhance protein identification and characterization in proteomics.

Main Methods:

• Developed an integrated platform for automated collision-, electron-, and photon-based fragmentation.
• Generated deep proteomics datasets using multi-enzyme workflows.
• Trained a unified Prosit deep learning model for spectral prediction.

Main Results:

• The Prosit model, integrated into FragPipe's MSBooster, increased protein identifications by >10%.
• Electron-induced and ultraviolet photodissociation achieved competitive identification efficiency with CID.
• Advanced fragmentation techniques provided superior sequence coverage and richer spectra.

Conclusions:

• Established a framework for routine application of advanced fragmentation techniques in proteomics.
• Demonstrated the potential of electron- and photon-based dissociation for comprehensive proteome analysis.
• Highlighted the utility of deep learning for unifying spectral prediction across dissociation methods.