Wayfinder
Wayfinder is a hypothesis referee.
Core trick
Generated gene-program-disease hypotheses are tested against CRISPRi knockdown data, returning supported, refuted, or untested verdicts with evidence at each hop.
Description
Wayfinder is a hypothesis referee. Literature-based discovery (LBD) overproduces hypotheses and tests almost none — the critique that has dogged the field for decades. Wayfinder answers it: it pairs LBD generation with a deterministic referee that tests each machine-generated gene→program→disease hypothesis against a genome-scale CRISPRi Perturb-seq resource, and returns a verdict with a receipt at every hop — including a confident, receipt-backed "no." A failed knockdown comes back untested, not a false negative: an artifact caught, not a discovery missed.
The whole loop ran inside Claude Science, whose independent reviewer model flagged and cut the project's own overclaims. On the Marson lab's CD4+ T-cell Perturb-seq data, 22,039 hypotheses were culled to a small set of receipt-backed candidates; the standout — NAB2→Th1/Th2→atopic eczema — is a connection the literature had never drawn (the eczema label is a genetic-association nomination). Measured as a predictor on held-out post-2016 links, the method returns a null at its primary metric — reported straight.
Why it matters: it makes machine-generated hypotheses triage-able and trustworthy — every claim carries an experimental receipt, and the system is willing to abstain and to refute.
README excerpt
# Wayfinder
**A hypothesis referee that says a confident *no*.** Every scientist is caught between two floods — over a
million papers a year and omics runs returning millions of measurements per sample. Wayfinder helps you
navigate the gap: it surfaces the data's implicit hypotheses with literature-based discovery, then **tests
each one against real experimental data and returns a verdict with a receipt for every hop** — *supported*,
*refuted*, or *untested (the knockdown failed)*. The library and the lab, on one bench.
> **▶ 3-minute demo video:** https://youtu.be/MbgojaAFfz0
>
> Built for **Built with Claude: Life Sciences** (Anthropic × Gladstone Institutes), July 2026 —
> **Researcher track ("a researcher who also builds").** The whole pipeline was reproduced natively inside
> **Claude Science**, Anthropic's scientific workbench.
## Why this exists
Three NIH reviewers killed the R01 behind this line of work (**1R01LM015392-01**) with one recurring critique:
> *"This application has the same major problem that has plagued all LBD work: it generates an enormous
> number of hypotheses, almost none of which ever get followed up."*
They were right. Literature-based discovery overproduces hypotheses and tests almost none. **Wayfinder is the
answer to that critique** — it puts hypothesis *generation* and *testing* on one bench, culls aggressively,
and is willing to say a confident, receipt-backed **NO**. Falsification, not confirmation, is the moat.
## The finding (a rece