Noozra
Two open-source tools, built with Claude, that catch expensive biology failures early — on public data, as decision-support (never a diagnosis).
Core trick
Twelve biology tools exposed over MCP let Claude agents screen protein sequences against human look-alikes and check variants against public databases.
Description
Two open-source tools, built with Claude, that catch expensive biology failures early — on public data, as decision-support (never a diagnosis).
The off-target copilot screens a therapeutic protein against its human look-alikes before the wet lab: anti-myostatin is 89.9% identical to GDF-11, so an antibody would hit both. It finds a selective epitope (identity to GDF-11 drops 89.9%→60%), checks whether animal models will engage it, and flags developability and immunogenicity liabilities — and it retrospectively reproduces 10/10 documented antibody cross-reactions. The clinical suite re-reads old genetic reports: it re-checks variants against current ClinVar, triages missense calls, flags drug–gene mismatches (CPIC), prioritizes genes from patient phenotypes (HPO), and reads gene constraint (gnomAD).
All twelve tools are driven by two Claude agents over MCP. Claude Code built the repo autonomously; five Claude agents researched the field in parallel; Claude Science ran the analyses and drew the figures. It never says "safe." 76 tests · one repo · MIT.
README excerpt
# In-Silico Safety & Decision-Support — Built with Claude: Life Sciences [](https://github.com/nochallenge/life-sciences-safety/actions/workflows/ci.yml) [](LICENSE) Two open-source tools that use **only public data** and **Claude agents** to catch costly, high-stakes mistakes *before* they happen — one before the **wet lab**, one before a clinician or patient acts on an **outdated or ambiguous genetic report**. Honest triage throughout: neither tool ever says "safe," and neither issues a diagnosis. Everything lives here, in one place, MIT-licensed. **At a glance:** 2 products · **12 tools driven by 2 Claude agents over MCP** · **76 automated tests** · off-target screen **validated 10/10** against documented antibody cross-reactions · every number computed live from public data. > **Why this exists.** This began as a personal design goal — two human-enhancement targets: > **myostatin (GDF-8)**, whose blockade builds **muscle strength**, and **BMP-2**, which drives > **bone formation and density**. Both sit in families packed with near-identical look-alikes > (anti-myostatin is **89.9%** identical to GDF-11; BMP-2 is **86.8%** identical to BMP-4), so > designing a *selective* binder for either forces you to confront off-target cross-reactivity > first. That safety problem is what this project ended up