A Billion-Scale Dataset for Proteomics AI

Standardized corpus combines mass spectra from 35 species and multiple instruments

October 22, 2026
3 min

The Analytical Scientist

At a Glance

Category	Detail
Condition	Proteomics data standardization and artificial intelligence for tandem mass spectrometry
Key Mechanisms	MassNet standardizes tandem mass spectra with peptide identifications and related metadata in MSDT format; XuanjiNovo uses spectra to predict peptide sequences de novo.
Target Population	Laboratory and medical researchers developing or evaluating proteomics models; the dataset includes samples from 35 animal, plant, and microbial species, with most data from humans and mice.
Care Setting	Research and laboratory settings; clinical diagnostic performance was not assessed.

Key Highlights

MassNet combines 1.54 billion tandem mass spectra from 648 PRIDE and iProX projects and 27,643 LC-MS/MS runs.
After reanalysis and deduplication, the dataset contained 548 million peptide-spectrum matches, 2.54 million unique peptides, and 133,422 proteins.
MSDT stores spectra alongside identifications and metadata and was approximately 82% smaller than the combined source spectral and search-result files.
In a 15-species benchmark, XuanjiNovo outperformed five other de novo sequencing models; performance varied across datasets and instrument platforms.
The resource excludes unidentified spectra and does not systematically include several modifications, including phosphorylation, glycosylation, and ubiquitination.

Guideline-Based Recommendations

Diagnosis

The study describes a research resource and de novo peptide-sequencing model; it does not provide clinical diagnostic criteria.

Management

MassNet and XuanjiNovo are available for research use; the study did not assess clinical diagnostic performance.

Monitoring & Follow-up

The study reports variation in model results across datasets and mass spectrometry platforms; some older-instrument data required additional model training.

Risks

MassNet includes only spectra confidently identified by FragPipe or Sage, potentially introducing search-related bias and excluding unidentified spectra.
The dataset covers only a limited set of common protein modifications; clinically relevant modifications were not systematically included.
The experimental rescoring model identified fewer matches than established FragPipe and Sage workflows.

Patient & Prescribing Data

No patient treatment population or prescribing data are reported. The resource contains proteomic data from 35 species, with most samples from humans and mice.

No treatments or prescribing recommendations were evaluated. The study did not assess clinical diagnostic performance.

Clinical Best Practices

Interpret model performance in the context of the dataset and mass spectrometry platform; the study reports variation across both.
Account for the resource's search-based inclusion criteria and limited systematic coverage of protein modifications when using MassNet.
Do not infer clinical diagnostic performance from the reported model benchmarks; the study did not evaluate it.