Could carbamazepine treat pancreatic cancer?

Robertium found 856 such cross-domain drug repurposing candidates by analyzing 167,145 biomedical papers. None have clinical trials yet. All have evidence chains in published literature.

Open-source. Reproducible. PubMed-validated.

167,145
papers ingested
24,285
hypotheses discovered
856
novel candidates
112
reproduce active trials
Example novel candidate
{
  "drug": "carbamazepine",
  "drug_domain": "epilepsy",
  "mediator": "KRAS",
  "outcome": "pancreatic ductal adenocarcinoma",
  "outcome_domain": "pancreatic_cancer",
  "score": 0.300,
  "evidence_papers": 4,
  "clinical_trial_status": "none",
  "literature_status": "novel_signal"
}

Live tracking leaderboard

We make our predictions public.

On 2026-05-12 we froze 1,618 drug repurposing hypotheses and committed them to a public, append-only ledger. Every month we re-query PubMed and ClinicalTrials.gov, then publish whether each one has gained literature support. No retrospective edits.

See the live tracking leaderboard →
14 Strong support
61 Partial support
138 New PMIDs since freeze

The Problem

Over 4,000 biomedical papers are published every day. Knowledge stays fragmented across journals, disciplines, and languages. The connections that would lead to new treatments often already exist in the published literature, but no human team can find them systematically.

Approach

Robertium reads open biomedical literature, extracts structured claims about drugs, genes, and diseases, and connects them into a knowledge graph. From this graph, it surfaces contradictions, gaps, and reasoning chains that point to new drug repurposing hypotheses.

The pipeline is domain-agnostic. Currently active across ten therapeutic domains: glioblastoma, epilepsy, ALS, Alzheimer's disease, pancreatic cancer, major depressive disorder, multiple sclerosis, type 2 diabetes, rheumatoid arthritis, and inflammatory bowel disease. New domains can be added in days, not months.

Read

Open biomedical literature from OpenAlex, PubMed, bioRxiv

Extract

Structured claims about drugs, genes, diseases via language models

Connect

Knowledge graph reveals contradictions, gaps, and reasoning chains

Open

Robertium is open-source under the MIT license. The code, the extracted claims, and the knowledge graph are all freely available.

This is intentional. Drug discovery infrastructure should belong to the scientific community, not to private platforms.

MIT License Open data Active development

Audit-grade transparency

Every hypothesis has a full provenance trace

Source papers, extracted claims, scoring code, embedding model, prompt version, and the exact pipeline run that produced the row. Downloadable JSON. No black boxes.

See an example trace → How it works