A large-scale unified deep learning model for peptide mass spectrum interpretation trained on multimodal data

2026-05-25 · Source: Nature Machine Intelligence · Field: Science & Research — Life Sciences & Biology, Artificial Intelligence & Machine Learning · Depth: Expert, long

Summary

pUniFind is a large-scale multimodal foundational deep learning model designed for peptide mass spectrum interpretation in proteomics. Trained on over 100 million open search-derived spectra, pUniFind unifies end-to-end peptide-spectrum scoring with zero-shot de novo sequencing by aligning spectral and peptide modalities through cross-modality prediction and other pretraining tasks. The model significantly outperforms traditional engines, showing a 42.6% increase in identified peptides in immunopeptidomics. For modification-rich de novo sequencing, pUniFind identifies 60% more peptide–spectrum matches despite a 300 times larger search space. It also recovers an additional 38.5% of peptides in regular de novo sequencing, including 1,891 peptides absent from reference proteomes but mapping to the genome. A deep learning-derived quality control module further boosts consistency with RNA-Seq evidence from 65.4% to 85.0%.

Key takeaway

For proteomic researchers and bioinformaticians analyzing complex mass spectrometry data, pUniFind offers a powerful new tool to enhance peptide identification and de novo sequencing. You should consider integrating pUniFind into your workflows, especially for studies involving immunopeptidomics or novel post-translational modifications, to uncover previously missed peptides and improve data consistency with genomic evidence. This can accelerate discovery in areas like biomarker identification and drug target validation.

Key insights

pUniFind unifies peptide-spectrum scoring and de novo sequencing, significantly improving proteomic analysis sensitivity and modification coverage.

Principles

• Multimodal pretraining aligns diverse biological data.
• Open scoring enhances interpretability and performance.
• Deep learning improves consistency with orthogonal evidence.

Method

pUniFind integrates end-to-end peptide-spectrum scoring with zero-shot de novo sequencing, trained on 100M+ spectra using cross-modality prediction and pretraining tasks. It offers two de novo workflows.

In practice

• Apply pUniFind for immunopeptidomics to increase peptide identification by 42.6%.
• Use for modification-rich de novo sequencing to expand search space 300x.
• Integrate its quality control module to align results with RNA-Seq.

Topics

• Proteomics
• Mass Spectrometry
• Deep Learning Models
• Peptide Sequencing
• Immunopeptidomics
• De Novo Sequencing

Code references

• pFindStudio/pUniFind

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

Related on AIssential

• Integration of alternative fragmentation techniques into standard LC-MS workflows using a single deep learning model enhances proteome coverage — A unified deep learning model and integrated MS platform enhance proteome coverage by leveraging diverse fragmentation techniques.
• Zero-shot de novo peptide sequencing with open posttranslational modification discovery — RNovA enables zero-shot open PTM discovery in de novo peptide sequencing with high accuracy.
• Zuckerberg's $500M AI biology swing — Large-scale data generation and advanced AI models are converging to revolutionize disease understanding and food science.
• Programmable RNA translation through deep learning-driven IRES discovery and de novo generation — An AI framework unifies IRES identification, optimization, and de novo generation for programmable RNA translation.
• 🔬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.
• Probabilistic modelling of single-cell bisulfite sequencing data with MethylVI — MethylVI disentangles biological signals from technical noise in single-cell bisulfite sequencing data.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Nature Machine Intelligence.