In memory of my little brother and best friend, John.
For John was created in honor of John Gegen. Researchers have spent decades building knowledge about osteosarcoma, often in separate labs, trials and journals. I’m leveraging frontier AI models to bring that work together, connect the dots between studies, and carry the connections to the people who can test them.
Not medical advice. Nothing here is a recommendation about any patient's care. Findings are leads for researchers to check, not conclusions.
What we found
Each finding carries a label for how solid it is: Confirmed (checked against the original source), Lead (supported, but needs a scientist to test it), or Hypothesis (a reasoned idea, not yet tested).
Kinase-inhibitor pills: active, but briefly
Patients free of progression about four months into treatment, relapsed osteosarcoma
Pooled from published trials and registry results, weighted by patients. The red line is the historical benchmark of about 12%.
Includes a kinase-inhibitor pillNo kinase-inhibitor pillHistorical benchmark
Show as a table
Treatment
Patients
4-month rate
What we're asking
If you work on osteosarcoma, or on any of the targets above, and something here is wrong or useful, please open an issue on GitHub. Corrections are as valuable as confirmations.
How this was made
Who: built by John Gegen’s brother, leveraging frontier AI models to gather, analyse and cross-check the public research.
From what: 33,242 PubMed papers mentioning osteosarcoma, 758 studies on ClinicalTrials.gov, NIH grant records (2015–2025), Europe PMC full texts, and open genomic and tissue-expression data.
Checked how: every headline claim was audited against its original source in two separate review passes run by AI models, plus a repo-wide consistency check, and the reports were corrected where these found errors. Each report ends with a change log.
Not yet: checked by any independent person or peer-reviewed. That is the next step, and we are asking for it. Auditors can start with the auditor's guide.
Open: most analyses can be rebuilt with the scripts in the repository; a few counts came from one-off database queries described in each report. Reports and data are licensed CC BY 4.0, code is MIT.