OpenOnco
Project news
Release notes, knowledge-base milestones and governance decisions. Every post is open for comment.
News
12 postsNormal skin is already full of "cancer" mutations
Sequencing physiologically normal eyelid skin found driver mutations — the same ones found in tumours — in roughly a quarter of cells, at about 140 per square centimetre. The skin was working fine. A driver mutation alone does not make cancer.
What NCCN category 2A actually means
2A is widely read as "weak evidence, weak recommendation." It is not. The panel agreement behind 2A is the same as behind category 1 — and 2A is what you get by default when no category is printed at all.
This section is new, and it is open for comment
Release notes and governance decisions now have a home on the site instead of living only in commit messages. Every post takes comments — with a few rules that a cancer-information site genuinely needs.
Decision-tree routing defects fixed, validator hardened
Auditing every treatment algorithm through the engine walker — rather than the loader — surfaced routing defects that were silently dropping selected indications from rendered plans.
Two doors — and why patients get a different one
The homepage now routes clinicians and patients to different places. Patients get help understanding a plan and preparing questions, not treatment recommendations — a line that has to hold for both safety and regulatory reasons.
An AI assistant can now call the engine instead of guessing
OpenOnco ships an MCP server, so an assistant answering an oncology question can route through the rule engine and return cited output rather than improvising a regimen.
Prevention plans, for people who do not have cancer yet
The charter was amended to cover people at elevated risk but without a diagnosis. The engine now produces a Prevention Plan alongside the Treatment Plan, and the first pilots are live on the site.
85% of our decision-tree conditions were unreadable to the engine
An audit of every treatment algorithm found that most branching conditions were written as English prose the engine cannot evaluate — and that 30% of algorithms therefore fell through to their default on every patient.
What landed in the knowledge base, 30 April to 11 May
A digest of one busy fortnight: haematology, thoracic and breast regimen waves, rare-tumour red flags, CIViC actionability backfill, and a batch of routing fixes.
Biomarker actionability moves from OncoKB to CIViC
A licence audit found OncoKB's terms incompatible with running OpenOnco as a free public resource. Biomarker actionability now builds on CIViC, which is CC0.
Why an LLM never picks the treatment
The project started out asking a language model to rank treatments. That approach is now archived, and the rule that replaced it shapes everything else: recommendations come from a versioned knowledge base, not a model.
Practising oncologists reviewed the engine's plans for the first time
The engine produced treatment plans for a real patient, and the oncologists who read them judged them strong. One case does not validate a system — but it is the first signal from outside the project's own scoring.