Community SDRF annotations for public proteomics datasets (ProteomeXchange and related accessions).
The SDRF specification lives in bigbio/proteomics-sample-metadata.
License: Apache 2.0 · Contributing: CONTRIBUTING.md · Agent context: llms.txt, AGENTS.md
Auto-generated from curated datasets/ on 2026-09-27T21:03:33Z. Sandbox drafts are excluded.
| Metric | Count |
|---|---|
| Accessions | 10,187 |
| SDRF files | 10,460 |
| Accessions with a declared template | 10,184 |
Samples (unique source name per file) |
316,072 |
Runs (unique comment[data file] per file) |
425,291 |
| Assay rows | 536,056 |
| Human contributors | 28 |
| AI agents (named fingerprints) | 5 |
| AI-assisted accessions | 10,187 |
| Unidentified agent | 2,554 |
| Multi-agent accessions | 340 |
| Distinct instruments | 152 |
| Median runs per accession | 12 |
| Accessions with modification parameters | 4,570 |
| ProteomeXchange coverage | 10,130 / 56,691 (17.9%) |
| PRIDE coverage | 9,969 / 41,770 (23.9%) |
Highlights: most common organism is Homo sapiens; 100,549 DIA assay rows; 97,260 TMT and 392,532 LFQ assay rows; 73 single-cell, 594 cell-line, and 594 metaproteomics accessions; sample-field completeness (applicable samples): disease 40%, age 13%; all 10,187 accessions are AI-assisted (28 human contributors, 5 named AI agents); identified fingerprints are mostly Cursor; 2,554 accessions have no vendor fingerprint (typical of Claude Code committed as the reviewer); 410 accessions have Codex evidence (codex/ PR branches); 340 accessions were touched by more than one agent (most common handoff Cursor → Codex); most common instrument is Q Exactive; most common modification is Carbamidomethyl; 17.9% of public ProteomeXchange datasets have a curated SDRF here; 23.9% of PRIDE projects are annotated.
Looking for a known-good SDRF to point a pipeline at? A short curated list is kept for
exactly that. Each entry passes both CI gates, maps every row to a real deposited run, and
carries enough sample metadata to exercise what usually breaks first — TMT channel maps,
cell-line identity, phospho-enrichment metadata and factor-value driven designs. Between
them they span five organisms, DDA and DIA, label-free, TMT and SILAC, and 112 to 5,798
rows. The standout is PXD030304: 949 cell lines with per-line sex, age, ancestry and
Cellosaurus accession across 5,798 individually mapped runs.
See docs/gold-standard-datasets.md for the list, what each one is good for testing, and how to fetch and validate them.
Some deposits have problems that annotation alone cannot fix. Examples are a 2 KB
metadata stub named .raw, or a PRIDE instrument field that contradicts the raw file.
These SDRFs are kept, not deleted, and are listed in
docs/known-issues.md by severity (critical, major, moderate,
minor), with the evidence for each, so pipelines can skip them and curators can follow up.
| Resource | URL |
|---|---|
| Specification | https://github.com/bigbio/proteomics-sample-metadata/blob/master/sdrf-proteomics/README.adoc |
| Public site | https://sdrf.quantms.org/ |
| Templates | https://github.com/bigbio/sdrf-templates |
Validator CLI (parse_sdrf) |
https://github.com/bigbio/sdrf-pipelines |
| Agentic toolkit | https://github.com/bigbio/sdrf-skills |
Files follow the pattern datasets/{ACCESSION}/{ACCESSION}.sdrf.tsv:
datasets/PXD000070/PXD000070.sdrf.tsv
datasets/MSV000078494/MSV000078494.sdrf.tsv
Additional .sdrf.tsv files may appear in the same folder when a project requires split designs.
Work-in-progress annotations live under sandbox/.
Move a folder to datasets/ and open a PR once it passes parse_sdrf validate-sdrf.
CI only validates datasets/; sandbox/ is exempt so drafts don't block merges.
Open a pull request to add or improve annotated SDRF files. See CONTRIBUTING.md for layout rules and review etiquette.
Use sdrf-skills as the primary toolkit. Key rules:
- Anchor every row in public evidence (PX page, submitted metadata, publication). Don't invent sample names or file names.
- Keep PRs small — one accession or a closely related batch.
- Run validation locally (
parse_sdrf validate-sdrf) before opening a PR. - Declare assistance in the PR description so reviewers can calibrate review depth.
For agent-specific instructions see AGENTS.md.
GitHub Actions runs parse_sdrf validate-sdrf on every PR and push touching datasets/**.
The validator is installed from bigbio/sdrf-pipelines main branch.
Re-run all checks manually via workflow_dispatch in the Actions tab.
- Dai C, Füllgrabe A, Pfeuffer J, Solovyeva EM, Deng J, Moreno P, Kamatchinathan S, Kundu DJ, George N, Fexova S, Grüning B, Föll MC, Griss J, Vaudel M, Audain E, Locard-Paulet M, Turewicz M, Eisenacher M, Uszkoreit J, Van Den Bossche T, Schwämmle V, Webel H, Schulze S, Bouyssié D, Jayaram S, Duggineni VK, Samaras P, Wilhelm M, Choi M, Wang M, Kohlbacher O, Brazma A, Papatheodorou I, Bandeira N, Deutsch EW, Vizcaíno JA, Bai M, Sachsenberg T, Levitsky LI, Perez-Riverol Y. A proteomics sample metadata representation for multiomics integration and big data analysis. Nat Commun. 2021 Oct 6;12(1):5854. doi: 10.1038/s41467-021-26111-3. PMID: 34615866; PMCID: PMC8494749. Manuscript
- Perez-Riverol, Yasset, European Bioinformatics Community for Mass Spectrometry. "Towards a sample metadata standard in public proteomics repositories." Journal of Proteome Research (2020) Manuscript.