Find the next use
for a drug that already works
Neolen reasons over a biomedical knowledge graph built from 35M+ PubMed abstracts and drug, gene, and trial databases to generate ranked repurposing hypotheses — then validates them against real hospital data in days, not years.
Phase 1 runs entirely on public data — no hospital partnership or PHI required to get value on day one.
Most drugs fail because the biological hypothesis was wrong
Target selection is the highest-leverage, highest-failure-rate decision in the entire drug pipeline — and it's made by humans reading literature, a process that cannot scale against a corpus growing by ~1.5M papers a year.
| Gap today | Neolen |
|---|---|
| Hypothesis generation is manual, bounded by human reading speed | Graph-scale inference across the full biomedical corpus |
| Connections across disciplines are missed (cardiology paper never read by neurology lab) | Graph traversal is discipline-agnostic by construction |
| Validating a hypothesis requires a new prospective study | Retrospective signal check against existing patient data in days |
| Cohort building takes weeks of manual chart review | Natural-language cohort query, in seconds |
| Hypotheses arrive as unexplained model outputs researchers won't trust | Every hypothesis ships with its full evidence path and source citations |
Two connected engines close the loop
The first half generates the hypothesis from open data. The second half validates it against real patients.
Discovery Engine
Builds a biomedical knowledge graph from public science and reasons over it to generate ranked hypotheses of the form: “Drug X, approved for Disease A, is mechanistically plausible for Disease B.” Every hypothesis ships with its full evidence path back to source papers.
Evidence Engine
Takes a hypothesis into real hospital data with natural-language cohort discovery:“Among patients who already take Drug X for Disease A, do those who also have Disease B show better outcomes than expected?”Runs inside the customer’s environment, behind the PHI boundary.
This has worked before — at the highest stakes possible
In 2020, BenevolentAI’s knowledge graph identified baricitinib — an existing rheumatoid arthritis drug — as a candidate COVID-19 treatment through literature-based inference alone. It was validated in trials and received emergency authorization.
An AI system generated a treatment hypothesis from public literature, and it turned out to be right. That is the ceiling Neolen is aiming at — which is why our Phase 1 exit gate is a blinded retrospective test: restricted to pre-2019 data, can the system independently rediscover baricitinib and 15–25 other known repurposing successes?
Built for the people who chase the hypothesis
Translational Researchers
Publishable hypotheses, fast literature synthesis, cohort access for retrospective studies.
Pharma R&D
New indications for in-portfolio assets, white-space analysis, de-risked 505(b)(2) candidates.
Rare Disease Foundations
Any plausible existing drug for a disease with no treatment and no commercial sponsor.
Biotech & In-Licensing
Undervalued repurposing opportunities worth in-licensing, sourced faster than manual scans.
The regulatory environment is pulling for this
A novel drug costs $1–2.6B and 10–15 years, largely because of safety testing. A repurposed drug has already cleared human safety for its original indication — under the FDA’s 505(b)(2) pathway, extensive preclinical work and Phase I can in some cases be bypassed entirely, moving straight to Phase II efficacy.
On 11 May 2026 the FDA opened a public docket explicitly soliciting input on its drug repurposing initiative, naming chronic disease, rare disease, neurodegenerative, metabolic, and substance use disorders as priority areas.
Explainable by requirement, not by exception
A ranked list with no reasoning will be dismissed by every domain expert who sees it. Every Neolen hypothesis renders its full evidence path — mechanism, corroborating papers, prior trial history, and safety signal — before it ever reaches a researcher's desk.
PHI boundary by architecture
The Discovery Engine touches only public data and runs as multi-tenant SaaS. The Evidence Engine touches patient data and runs inside your environment — only aggregate statistics ever leave.
Negative evidence filtering
Candidates already tried and failed in ClinicalTrials.gov, or contraindicated per FAERS, are actively down-ranked before they ever reach you.
Signals, not proof
Real-world evidence outputs are labeled as hypothesis-supporting signal, never as proof of efficacy. Only trials establish causation.
Bring a hypothesis to your PI, board, or grant committee — with the evidence attached
We're recruiting 2–3 rare disease foundations and academic research groups as founding design partners.