OncoMatch

Put forward a hypothesis and you’ll be assigned a team: OncoMatch will find people at the National Institute of Public Health who have a proven track record in this field, as evidenced by their publications.

You write a research hypothesis. In no time at all, you’ve got a team.

We tried: “Targeted therapy for glioblastoma via FAP; we need tumour samples from patients, a model of brain metastases and spatial transcriptomics.” OncoMatch proposed a team spanning Prague and Brno, from two different groups at the National Institute of Public Health. For each member, it explained why they were chosen, what they would bring to the team, what was still missing, and how to get started. All with a reference to a specific article.

Why the NÚVR needs this. 65 groups, 3 hubs, 17 institutions. Finding out who has which model, technology, samples or patients currently involves phoning round people we know. Collaboration that would make the most sense will not emerge so easily, because people are unaware of one another’s existence.

What makes OncoMatch different

It is not a chatbot that makes things up. Every recommendation is based on a documented publication (DOI). The system discards any output lacking evidence.
It knows what everyone has actually done. We extract roles from more than 1,300 author contributions (CRediT): who designed the study, who provided the patient samples, who analysed the data. These are grouped into the categories Research / Biology / Technology.
It looks for complementary work, not competitors. It breaks down a hypothesis into requirements and finds someone to fulfil each one.
It identifies what is missing from the network. It openly highlights gaps, rather than offering the closest but unsuitable match.
It is currently running at oncomatch.eu.

Data. Nearly 2,000 researchers (both team leaders and members), over 4,600 publications (NÚVR, OpenAlex, PubMed, RIV) and SCImago journal quartiles. In addition, over 1,600 oncology clinical trials in the Czech Republic, including comprehensive cancer centres, dozens of research infrastructures and over 100 BBMRI biobank collections. Public data only; no patient data.

Technology. The LLM breaks down the hypothesis into requirements → vector search in InterSystems IRIS Vector Search (multilingual-E5) → deterministic reordering (similarity, role, citations, recency) → the LLM assembles a team and an explanation, which we verify against the data.

Next steps. A pilot with the National Institute for Cancer Research (NÚVR) and a demonstration at the Czech Annual Cancer Research Meeting in Olomouc (23–25 November 2026). The same system can be set up for EATRIS and other networks: their public data is all that is needed.