oncologydata.ai

Public oncology datasets

Find studies, check their source metadata, and retrieve the files you need.

Matches describe study metadata, not verified patient eligibility. Use quotes for phrases and a minus sign to exclude a term.

Repository search contacts GEO, GDC, and cBioPortal and searches the verified CELLxGENE catalog snapshot. Use public research terms. Up to 20 matches per source; results are not exhaustive.

Loading catalog...

Cohort basket

Each study is saved in its own folder. This does not harmonize measurements or establish independent patients.

Download code

Python 3.10+ is required for all tabs. R and shell run the same downloader. Files are not extracted or harmonized. The default total download limit is 5 GB.

Run Python or shell with --list to inspect the file manifest first. Increase --max-bytes only after reviewing it.

AI-assisted selection

Optional: use your provider key to rank the current results. The query and up to 30 result descriptions go directly to that provider. Provider charges apply.

Keys stay in memory for this tab and are cleared on reload. No key is needed for catalog or repository search.

No key loaded.

Model suggestions are not verification of biomarkers, clinical fields, or demographic subsets. Check those in the source.

Using the index

What does a matching study establish?

A match means its metadata mention your terms or a cancer-site synonym. It does not establish that individual patients carry a mutation or meet an eligibility criterion. Counts retain their source units: patients, samples, and cells are not interchangeable.

How do I download the data?

Choose a verified record, select its retrieval scope, then open its code. All tabs use a standard-library Python downloader. It resolves actual files, records checksums and source URLs, and reports failures. GEO supplementary archives may contain mixed assays. Review the manifest and source before analysis.

Why do some entries offer only a source link?

Repository directories and records awaiting verification have no automatic download recipe. Follow the source link to choose a study and check permissions. We do not replace unavailable data with a different cohort.

Can I combine studies in the basket?

The basket downloads into separate folders. Related accessions may reuse samples. Even without a known relationship, studies may overlap. Batch correction, feature alignment, and patient deduplication require a study-specific analysis.

How do I cite and reproduce a search?

Export the selection JSON to retain the query, filters, time, metadata, verification dates, and source publications. Cite the originating study and accession. Keep the downloader's manifest with the retrieved files. Live source metadata and files can change.

Are public data always unrestricted?

No. Public metadata can describe controlled files. GDC recipes select open files. Other repositories retain their own registration, licensing, and data-use requirements. Follow their terms; this index does not grant additional rights.

How does the AI key work?

Validate a key to load models directly from the selected provider. It remains in this tab's memory. A successful model-list request does not guarantee that every model supports ranking. Request errors are shown; heuristic results are never disguised as model output.

Source coverage

These counts describe the local catalog, not each repository's full holdings. Directory entries are links to repositories, not individual datasets. GEO refresh covers existing accessions; live search can find more.

Privacy and source terms

The catalog is a static public metadata index. It does not host patient data. Typing searches locally. Repository search contacts GEO, GDC, and cBioPortal, which receive ordinary request information such as your IP address. GEO receives the query; GDC and cBioPortal return metadata that is searched locally. CELLxGENE search uses the verified catalog snapshot because its API does not permit direct browser requests.

AI ranking sends your question and candidate metadata directly to the provider you choose. Its fees and privacy terms apply. Keys are held only in this tab's memory. Basket selections and display preferences are saved in your browser. Older saved API keys are removed on startup.

No third-party analytics or runtime scripts are loaded. Hosting providers may keep access logs. Contact opens your email application; no contact form processor receives your message.

Use only public research terms. Do not enter patient identifiers or confidential study information. Dataset ownership, licenses, registration, and access requirements remain with the original sources. Verification describes retrieval of source metadata, not independent scientific validation.

Independent research software. No repository endorsement or clinical suitability is implied. Code is provided under the repository license; underlying data retain their original terms.