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

2026-03-23 · Source: Machine learning : nature.com subject feeds · Field: Science & Research — Life Sciences & Biology, Mathematics & Computational Sciences, Research Methodology & Innovation · Depth: Expert, extended

Summary

A new integrated mass spectrometry platform and a unified deep learning model, Prosit_2025_intensity_MultiFrag, significantly enhance proteome coverage by incorporating alternative fragmentation techniques beyond traditional collision-induced dissociation (CID). Researchers developed an Orbitrap-Omnitrap hybrid instrument capable of automated collision-, electron- (ECD, EID), and photon-based (UVPD) fragmentation. They generated extensive multi-enzyme datasets to train the Prosit model, which predicts spectra across all these dissociation methods. Integrated into FragPipe's MSBooster module, this publicly available model increased protein identifications by over 10% on average for both data-dependent (DDA) and data-independent acquisition (DIA) across all techniques. The study demonstrates that electron-induced and ultraviolet photodissociation, which yield richer, more informative spectra, achieve identification efficiency competitive with CID while providing superior sequence coverage, establishing a framework for routine application of advanced fragmentation in proteomics.

Key takeaway

For AI Scientists developing proteomics pipelines, you should integrate the Prosit_2025_intensity_MultiFrag model into your workflows, particularly for analyzing data from electron-induced and ultraviolet photodissociation. This model, available via Koina and integrated into FragPipe's MSBooster, significantly boosts protein identification rates and sequence coverage, especially for challenging proteoforms. Adopting these advanced fragmentation techniques and the associated deep learning tools will enable more comprehensive and confident proteome characterization, moving beyond the limitations of traditional CID methods.

Key insights

A unified deep learning model and integrated MS platform enhance proteome coverage by leveraging diverse fragmentation techniques.

Principles

• Alternative fragmentation methods yield richer spectra.
• Deep learning can unify diverse spectral prediction.
• Rescoring improves identification accuracy and recovery.

Method

An Orbitrap-Omnitrap hybrid instrument was developed for automated multi-modal fragmentation. Large-scale multi-enzyme LC-MS data trained a unified Prosit deep learning model for pan-fragmentation spectrum prediction, integrated into FragPipe's MSBooster for rescoring.

In practice

• Utilize Prosit_2025_intensity_MultiFrag for enhanced peptide ID.
• Employ alternative fragmentation (UVPD, ExD) for complex proteoforms.
• Integrate multi-enzyme digests for broader proteome coverage.

Topics

• Deep Learning Proteomics
• Mass Spectrometry Fragmentation
• Prosit Model
• Proteome Coverage Enhancement
• Peptide Spectrum Prediction

Code references

• wilhelm-lab/oktoberfest
• wilhelm-lab/koina
• Nesvilab/MSBooster
• wilhelm-lab/Prosit_multifrag

Best for: AI Scientist, AI Researcher, Machine Learning Engineer, Research Scientist

Related on AIssential

• A large-scale unified deep learning model for peptide mass spectrum interpretation trained on multimodal data — pUniFind unifies peptide-spectrum scoring and de novo sequencing, significantly improving proteomic analysis sensitivity and modification coverage.
• How to Accelerate Protein Structure Prediction at Proteome-Scale — Accelerated computing and optimized workflows enable proteome-scale prediction of protein complex structures.
• EnzymeTuning improves enzyme-constrained metabolic modeling and proteome abundance prediction through deep learning — EnzymeTuning, a GAN-based framework, optimizes *k*cat values to enhance metabolic modeling and proteome prediction accuracy.
• DDA-BERT: end-to-end training for data-dependent acquisition mass spectrometry-based proteomics — DDA-BERT is a transformer-based model that significantly improves peptide identification in DDA proteomics through end-to-end deep learning.
• SpecGP as a transformer-based model for predicting energy-adaptable structural spectra of glycopeptides — SpecGP uses a transformer architecture to predict glycopeptide spectra and retention times, enhancing isomeric discrimination and identification.
• 🔬ESM: The Bitter Lesson is Coming for Proteins - Alex Rives, BioHub — Scaling protein language models with vast, diverse data reveals emergent biological capabilities and enables programmable design.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine learning : nature.com subject feeds.