The world's largest collection of paediatric cancer cell lines.
The Childhood Cancer Model Atlas, housed at Hudson Institute's Centre for Cancer Research, leverages the world's largest collection of paediatric cancer cell lines and AI-powered analytics to drive innovative research and clinical translation. It includes more than 400 high-risk paediatric cancer cell lines, established through the collaboration of nearly 50 leading cancer research institutes, universities and academic medical centres worldwide.
The CCMA is open to every paediatric oncologist and childhood cancer researcher. Using it, you can have potential treatments tested and analysed with cutting-edge techniques, and draw on shared big data to find new therapies and biomarkers faster.
The connector adds a third way in, next to the dashboard and the analytical portal. Use the portal to browse and plot. Use the connector when you have a specific question and want the answer, with its source, inside the tool you are already working in.
Ask in your own words
Name the cohort, gene or compound you care about. You do not need to know the table names, the assay codes or how the release is structured.
Cross assays in one question
Drug sensitivity, CRISPR dependency, variants, expression and methylation sit together, so one question can span them instead of three exports and a join in R.
See what was never tested
A model an assay did not cover is reported differently from one that was tested and showed nothing, so a gap in coverage does not read as a negative result.
Paediatric by default
Seventeen adult high-grade glioma lines share the release. They stay out of your results unless you ask for them, so a paediatric answer is never quietly pooled with adult data.
Questions researchers put to it in a first session.
Paste any of these into a connected assistant. It chooses the query, keeps to the paediatric cohort, and returns the ranked result with its coverage caveats.
{{ q.text }}
{{ q.note }}
Results describe cell line models, not patients. They are pre-clinical, useful for forming a hypothesis, and not a basis for treatment decisions.
Set it up once in the tool you already use.
Hudson hosts the service, so there is nothing to install. You add one address to your AI tool and it stays available for every session after that. Choose your tool below.
- {{ s }}
{{ activeBlockText }}
Access is read-only. If your institution restricts custom connectors, your IT team can approve this one address for the whole group.
The connector fetches the data. The skill makes Claude read it like a scientist.
Same question, same rows, two very different answers. On the left, the connector on its own. On the right, the connector with hudson-ccma installed.
"How many DIPG lines in the atlas carry an H3K27M mutation?"
Four DIPG lines carry an H3K27M mutation.
Correct as far as it goes, and easy to misread. Four out of how many? Were the rest tested? And a search on K28M, the nomenclature half the field now uses, would have returned nothing at all.
Four of the 11 DIPG lines with variant data carry H3K27M, also written K28M. Three further DIPG lines are in the atlas but were never sequenced, so their status is unknown rather than negative. ccma:9f2a1c4b7e03
A denominator, an explicit unknown, both spellings of the variant, and a token you can resolve back to the source rows.
Coverage denominators
Every count arrives against the number of models actually tested, never as a bare figure.
Not tested, or no match
An untested model is reported as unknown, so absence of data never reads as a negative result.
H3K27M and K28M
The two numbering conventions are treated as one variant, whichever you type.
Citation tokens
Answers carry the token that resolves to the source rows, ready for your working notes.
One file, no dependencies. Works with the connector you set up above.
Download the skillVersion-controlled hosting on GitHub is coming. For now this is a direct download served alongside this page. If the download does not start, email info@hudson.org.au and we will send the file.
Add the ccma connector, following the Connect section above.
In Claude, open Settings, Capabilities, Skills, and upload the downloaded file.
Ask your question. Claude applies the skill whenever it calls the ccma queries.
The skill changes how results are explained, not what the connector can reach. Everything in Scope and limits still applies.
Check any number before it goes in a paper.
Answers carry a short reference token. Ask your assistant to resolve it and you get the source rows behind the figure, down to the file, line and hash. The same token returns the same rows next year, so a reviewer can repeat the check.
"Which Ewing sarcoma lines in the atlas are most sensitive to CDK4/6 inhibition, and how strong is the effect?"
The matching lines ranked by z-score, a note of which lines the drug screen never covered, and a reference token.
Cite the CCMA release DOI as you would any dataset, and keep the token with your working notes.
Eleven read-only queries over the published release.
You never have to name one of these. They are listed for reference, and for anyone who wants to know exactly what a question can reach.
{{ t.desc }}
What the data is, and what it is not.
Cell line models, not patients
Every record describes a model. Nothing in the release identifies a person, and no clinical record is involved.
Pre-clinical evidence
Results support hypothesis generation and study design. They are not a treatment recommendation for any individual patient.
Read-only access
The service reads a fixed, DOI-versioned public release. There is no way for a question to alter the atlas.
A subset of the collection
The connector covers the 182 models in the current published release. The wider collection of more than 400 lines stays available through the analytical portal.
Column meanings are flagged
Where a column's meaning is verified against the source publication, it says so. Where it is unconfirmed, it says that too.
Open licence
The release is published under CC BY 4.0, DOI 10.17632/rnfs539pfw.3. Cite it as you would any CCMA dataset.
Questions researchers ask first.
Do I need to know how to code?
No. You type a question the way you would ask a colleague. The assistant works out which query to run and reports what it did, so you can check the reasoning as well as the number.
Which data is covered?
Drug screening z-scores, CRISPR knockout dependencies, small variants and copy-number changes, gene expression as log2(CPM+1), and Illumina methylation beta values, for the 182 paediatric models in the current release.
Can I use results in a publication?
Yes. The release is CC BY 4.0, so cite the DOI. Resolve the reference token for each figure you quote and keep the source rows with your working notes, so any number can be traced later.
Is this real patient data?
No. Every record describes a cell line model. Findings are pre-clinical and are not a treatment recommendation.
Where does my question go?
Your AI tool sends the query to the hosted service and the service returns rows from the public release. The service runs no AI model of its own and holds no keys, so nothing is inferred or stored on Hudson's side beyond ordinary request logs.
Does it replace the analytical portal?
No. The portal is still the place to browse, plot and explore. The connector is for specific questions, quick checks and pulling figures into something you are already writing.
My AI tool is not listed. Will it work?
Any tool that can connect to a remote MCP server over HTTPS works. Use the Other tab and paste the address where your tool asks for one.
Who do I contact about the atlas?
For research questions about the CCMA and the Childhood Cancer Program, contact Hudson Institute's Centre for Cancer Research at info@hudson.org.au.
Hudson Institute of Medical Research is affiliated with Monash Health and Monash University and a partner of the Monash Health Translation Precinct.
Developed by Ippon Australia for Hudson Institute of Medical Research.
Hudson Institute of Medical Research acknowledges the Bunurong People of the Kulin Nation as the traditional owners of the land on which we work. We pay our respects to Elders past and present. ABN 48 132 025 024.