PRIDE Proteomics Archive data

It is the world's leading public repository for mass-spectrometry-based proteomics, delivered as queryable rows. PRIDE - PRoteomics IDEntifications Database - operates at EMBL-EBI and is a founding member of the ProteomeXchange Consortium, the arrangement under which major journals require proteomics data deposition. That mandate is why the corpus grows the way it does: 40,809 submitted projects as of August 2026, roughly 500-900 new submissions a month, and monthly submitted-data volumes of 50-100 TB.

Each project is a study, keyed by a PXD-prefixed accession and stamped with a DOI. It carries a title, a project description, free-text sample-processing and data-processing protocols, and controlled-vocabulary annotations covering organisms, organism parts, diseases, instruments, quantification methods, experiment types, software, countries and submission type. Beneath it sit the file records: raw instrument output, search-engine results, peak lists and processed result files, each classified by category with sizes and checksums. Read as browser tabs this dies in folders; resolved into rows keyed by accession, which shotgun-proteomics studies used a Q Exactive with TMT labeling on human tissue becomes a filter, not a literature review. Get a sample of this dataset and inspect real records before committing pipeline time.

Three real records captured during the August 2026 pass - the archive's oldest current project, its newest observed submission, and one file record from beneath a project:

# Row 1 — the oldest project in the current archive, complete record
accession           : PXD001357
title               : Direct evidence of milk consumption from ancient human
                      dental calculus, St Helena
doi                 : 10.6019/PXD001357
submissionDate      : 2014-10-15
submissionType      : COMPLETE
organisms           : Homo sapiens (human)
instruments         : LTQ Orbitrap Elite | Q Exactive
experimentTypes     : Shotgun proteomics
projectTags         : Technical | Metaproteomics
totalFileDownloads  : 29298

# Row 2 — the newest submission observed in the August 2026 research pass
accession           : PXD082855
title               : Mapping the GW182 CIM1 binding site on CNOT1(800-999)
                      by HDX-MS
submissionDate      : 2026-08-19

# Row 3 — a file record hanging beneath a project
fileCategory        : Search engine output file URI
fileName            : F116783_JH8_QE.dat
projectAccession    : PXD001357

Read what the spread proves. Twelve years separate those two projects, yet the record shape never moved: accession, DOI-bearing title, dated submission, CV-annotated science. The first row shows why the annotations earn their keep - two instruments named precisely enough to compare, an experiment type drawn from a controlled vocabulary, and a download counter (29,298) that no citation metric captures. The third row shows the grain beneath the grain: a project is not one blob but a classified set of files, and fileCategory is what tells a spectral-library builder from a reanalysis team which layer to take. Example rows are illustrative of the record shape; field-level detail follows below.

Nineteen core fields cover project and file records, every definition checked against live records during the August 2026 research pass. They do three jobs.

Identity and dating: accession anchors the row on the PXD identifier every downstream resource cites, doi gives it a permanent citable handle, and submissionDate plus publicationDate place it on the timeline. submissionType grades completeness - COMPLETE, PARTIAL or PRIDE - which is a ready-made filter for whether a deposit holds everything a reanalysis needs.

Scientific context: the controlled-vocabulary block that makes this corpus joinable - organisms from NEWT taxonomy, instruments from the MS ontology, experimentTypes and quantificationMethods from PRIDE's own term lists, plus submitter keywords and projectTags. Because these arrive as vocabulary terms rather than free text, an instrument comparison across decades is a GROUP BY instead of a string-matching archaeology dig.

Methods and weight: the two protocol texts preserve how samples were prepared and how spectra were processed - the reproducibility record papers summarize away - while fileCategory, fileName, fileSizeBytes and totalFileDownloads describe and size the file estate beneath each project. Entries flagged in the footnote fold under additional fields on request.

Beyond the nineteen-column core, four extensions extend the same universe sideways; each gets mapped to your use case at sampling:

• Checksums and public location records on files - present on file rows during verification; they ship when your pipeline needs integrity checking or mirroring.
• Organism-part and disease annotations - further CV-param blocks carried on subsets of projects, folded in when clinical slicing matters.
• Protein identification records - the protein-level layer beneath projects, joined on by accession when identifications rather than studies are the unit of work.
• Reanalysis links and SDRF metadata exports - available where submitters provided them; SDRF availability in particular varies project by project, so it is confirmed per slice rather than promised blind.

Name the ones your models need when you request a sample and the extract comes back cut to exactly that shape.

Geography: global by construction. Laboratories worldwide deposit through ProteomeXchange routes, and a country facet rides on every project - so contribution geography is a queryable column rather than an inference from author affiliations. Organism, more than country, is usually the axis that matters: coverage tracks what the world's mass spectrometers have been pointed at.

Temporal: the current archive runs from PXD001357, submitted 2014-10-15, forward through present-day submissions - the newest observed during the August 2026 pass landed on 2026-08-19. One structural honesty note: material older than the ProteomeXchange era sits in a separate legacy resource outside this feed, so decade-spanning backtests should say so upfront. Freshness is not a constraint on your side either way - deliveries land daily, weekly, or hourly regardless of the underlying cadence.

Granularity: the project is the unit - a study with its protocols, annotations and DOI - and two finer layers hang beneath it: file-level records classified by category and sized in bytes, and protein identification records exposed as a further stratum. Fix the grain before scoping a request, because mixing studies with their constituent files inflates counts fast.

Scale check against Datadory's wider catalog - 1,744 datasets across 159 viable industries, average quality score 7.81 - this slice scores 8/10 with field definitions verified during research, and its growth rate means any snapshot understates next quarter's holdings.

API, files, or your warehouse. Daily, weekly, or hourly.

You pick the channel and the cadence; the field dictionary above travels unchanged through all three. Rows arrive flattened - one project per row, file records normalized beneath, CV terms kept as clean strings - so joining against your own target list, instrument inventory or disease panel is a join statement rather than a parsing project. Cadence changes are a settings conversation, not a re-integration, and a sample cut to your named organisms, instruments or date range comes first either way.

• Biomarker discovery panels - filter by organisms, disease annotation and quantificationMethods to assemble candidate cohorts before reading a single paper; the TMT-versus-label-free split alone collapses thousands of studies into the comparable subset.
• Instrument fleet analytics - the MS-ontology instruments block turns 40,809 projects into a market-share history of mass spectrometers: which platforms ran which experiment types, when, at what volume. Vendors and procurement teams price capacity decisions off that curve.
• Method reproducibility audits - sampleProcessingProtocol and dataProcessingProtocol text plus the submissionType completeness grade let reviewers check whether a paper's stated workflow matches what was actually deposited. Discrepancy becomes a query, not a peer-review hunch.
• Spectral library building - fileCategory separates raw instrument output from search-engine results and peak lists, so library builders pull the right stratum without downloading archives blind.
• Metaproteomics and ancient-protein research - projectTags isolates technical and metaproteomics deposits (dental-calculus milk proteins included) that generalist catalogs bury.
• Research-trend intelligence - submissionDate joined with experimentTypes yields a monthly-resolution series on what the field is actually measuring; see competitive intel workflows.

Ranked by relevance in Datadory's persona tagging:

1. Bioinformaticians & computational biologists - accession-keyed project and file records that join outward on DOIs and NEWT taxonomy without fuzzy matching, so reproducing a paper's inputs is a filter instead of a correspondence chain.
2. Data scientists & ML engineers - tens of thousands of labeled projects (organism, instrument, method, tag) as supervision for retrieval and cohort-matching models; the data scientists working in biotechnology briefing goes deeper.
3. Competitive intelligence & product teams - instruments and software annotations read which platforms dominate real depositions when judging vendor positioning; see developers building in biotechnology for the product angle.
4. Journalists, academics & students - DOI-carrying records citable at source, with totalFileDownloads as a ready-made impact story no citation count tells.
5. Investors & diligence teams - submission-volume trends by method and instrument as a demand signal on the proteomics toolchain before the term sheet moves.

The Biotechnology shelf splits the proteomics job by vantage point, and this record owns the deposition-evidence layer. Human Protein Atlas says where a protein appears in the body; the archive here supplies the mass-spec evidence that a peptide signature was actually measured - pair them and a tissue-restriction claim gains its instrument-grade receipt. UniProt provides the protein identities every identification resolves against, and RCSB Protein Data Bank supplies the experimental structures a validated interaction must fit into. European Nucleotide Archive covers the genomics side of the same mandate-and-deposit world, with a matching accession-first record grammar. GWAS Catalog adds the trait associations worth checking when a proteomic hit needs genetic corroboration. All of them, ranked and cross-linked, sit in the best biotechnology datasets shortlist; the pooled biotechnology data hub holds the full view, and the PRIDE (EMBL-EBI) source profile covers the repository itself.