judegomila/OnCo

Live: https://onco.cc

OnCo is an open repository for cancer. The whole corpus, every page, source, drawing, translation and script, is in this repository under an open licence (CC BY-NC 4.0 for the data, MIT for the code), built in public, corrected through issues and pull requests, and free for patients, clinicians, researchers, companies and AI agents to read and reuse. If you know something the site does not, open an issue or a pull request: the gate runs the same checks on every contribution.

OnCo is a public, cited knowledge graph of oncology, aiming at total information dominance on cancer: every cancer, front, technology, target, biomarker readout, treatment and test, company, institution, person, pathway, trial, key paper, journal, pairing, roadmap, bottleneck and idea, one page each, with a plain-English TL;DR before the technical layer and links in every direction. Readers start in a Global view and pick their country, so "approved" means their regulator's verdict. It exists so that a patient, a carer, a clinician, a scientist, an investor or a policymaker can walk from any object to everything connected to it and see the state of the art, the history, what is being done about each problem, and what is coming.

The website is one surface. The same corpus ships as:

• the site (https://onco.cc): tables with filters, tooltips on every technical term, molecule and target renderings, animated schematics, a graph explorer, Ask OnCo question answering, a living-with-cancer area (side effects, symptoms, second opinions, hair loss, complementary approaches graded by evidence), coverage of what insurers and the NHS pay for, and a completeness dashboard that states how much of the known world each kind covers;
• a static JSON API under https://onco.cc/api/v1/ with CORS, per-entity Markdown context files for language models, Atom feeds and RDF triples;
• an MCP server and a command-line tool (packages/), so agents and scripts can search, fetch, compare and ask over the corpus;
• browser WebMCP tools for read-only record search and lookup in supporting browsers (implementation and testing);
• eight interface languages with translated TL;DRs where a translation exists.

Work in progress. Verify at source. Every fact is being built and checked in the open and may be incomplete, out of date or wrong. Nothing here is medical advice. Do your own research and check anything that matters at its primary source, which every page links.

Current counts live at https://onco.cc/about/ and print from npm run validate. This file deliberately carries no numbers, so it never goes stale.

Captured from the built site on 17 September 2026. Click any image to open the live page.

Every treatment with a small-molecule structure shows the exact molecule as a slowly rotating wireframe; antibodies, cells, vaccines, devices and tests show an explained placeholder. Fronts, technologies, cancers, targets and glossary categories carry hand-drawn animated schematics. Readers pick their country in the header, and "approved" then means their regulator's verdict, with other regions shown as flags.

src/data/            The corpus: TypeScript records, one array per kind or theme
  index.ts           ALL_INPUTS: every array that becomes part of the graph
  cancers.ts, drugs.ts, targets.ts, technologies.ts, companies.ts, institutions.ts, ...
  spikes/            Per-cancer deep dives merged into the cancer records at load time
  people/, key-papers/, ideas-waves/, institutions/   Multi-file kinds, each with an index
  regional-approvals.ts, payloads.ts, resistance.ts, biomarkers.ts, ...   Structured side data
src/lib/schema.ts    Zod schemas for every kind, KIND_META (routes, labels), REL_FIELDS
src/lib/graph.ts     Builds the graph: validates ids, resolves references, derives backlinks
src/lib/nav.ts       Navigation groups and every top-level route
src/app/             Next.js App Router pages; src/app/[kind]/ renders every index, src/components/EntityDetail.tsx every object page
src/components/      UI: EntityBrowser (the shared filterable table), Wireframe3D and schematics, Tip (tooltips), icons, maps
scripts/             Fetchers and checks: structures (PubChem/RCSB), logos, OpenAlex, GLOBOCAN, trials, papers, factcheck, audit, provenance, API build
public/              Self-hosted structures, logos, portraits, snapshots, the static JSON API under /api/v1/
packages/            onco (CLI) and onco-mcp (Model Context Protocol server) over the API
docs/                GAPS.md (what is still missing) and design notes

1. Find the kind's file under src/data/ (or the right sub-file for multi-file kinds) and add or edit the record. Every record needs an id, name, tldr (plain English, one or two sentences), summary, asOf, and at least one primary links entry.
2. Reference other objects by id in the relationship fields (cancers, drugs, targets, technologies, trials, terms, people, bottlenecks, keyPapers, related, and so on). Backlinks are derived, so link in one direction only.
3. Run npm run validate. It rejects duplicate ids, dangling references and malformed records, and prints the counts per kind.
4. Run npm run typecheck && npm run lint && npm test && npm run build.
5. Open an issue first (the site's forms prefill it); pull requests that change data reference the triaged issue, so every change passes the same safety check.

New kinds need: a schema in src/lib/schema.ts (plus KIND_META, and the rel field in Base and REL_FIELDS if objects can point at it), a data file spread into ALL_INPUTS, a case in buildBrowser (src/app/[kind]/page.tsx) for the index table, a case in kindTabs (src/components/EntityDetail.tsx) for the page, a colour in src/lib/text.ts and GraphExplorer, and a nav entry.

GitHub Actions run the checks on every push and open weekly pull requests with refreshed snapshots.

• One page per object, links everywhere. If it can be a page, it is a page.
• Plain English first. Every object opens with a TL;DR a newcomer understands; the technical layer follows. Technical terms are glossary objects and get hover explanations wherever they appear.
• Cited, dated, honest about gaps. Every record links a primary source, carries the date it was last checked, and says "no data" rather than guessing.
• Nothing stale by design. Counts and dates are computed at build time from the corpus, never written into copy.
• Open for good. Code is MIT. Data is copyright OnCo, free for individual and educational use under CC BY-NC 4.0 with attribution; commercial use must contact us to pay for the data (see LICENSE-DATA). Logos remain their owners' trademarks.

What changed, release by release, is in CHANGELOG.md and rendered live at onco.cc/changelog. The Unreleased section is regenerated from the commit log every time the site ships (npm run changelog, also run by npm run build), so it is never behind the live site. Version sections are written by hand when a release is cut.

• Cite: CITATION.cff holds the citation for the software and, under preferred-citation, for the corpus as a dataset; GitHub's "Cite this repository" button reads it. Each tagged release attaches the corpus (JSON, NDJSON, CSV, JSON Schema, OpenAPI, Markdown context files) at https://github.com/judegomila/OnCo/releases, and Zenodo archives releases with a DOI once the repository is linked there; the DOI goes into CITATION.cff when issued.
• Reuse: data under CC BY-NC 4.0 (LICENSE-DATA): free for individual and educational use with the attribution line "Data from OnCo (onco.cc)" and a link to https://onco.cc; commercial use must contact us to pay for the data (issue labelled "licensing", or the contact on https://onco.cc/about/). Code under MIT (LICENSE). Every export, API file and CLI result carries the same attribution line.
• Machine access: https://onco.cc/api/ (endpoint list), https://onco.cc/api/v1/openapi.json (OpenAPI 3.1), https://onco.cc/llms.txt and https://onco.cc/llms-full.txt (for language models), npx onco and npx -y onco-mcp (docs/ACCESS.md). Distribution plan and press kit: docs/DISTRIBUTION.md, docs/press-kit.md.

See CONTRIBUTING.md, CODE_OF_CONDUCT.md and SECURITY.md. No personal or patient data belongs in this repository; only public knowledge about the science, the products and the organisations.

Data and structures from PubChem, RCSB PDB, Wikimedia Commons and Wikidata, OpenAlex, GLOBOCAN (IARC), ClinicalTrials.gov, Europe PMC, openFDA and the publishers and institutions linked on each page.