Mutation-aware drug-repurposing triage

When a tumor mutates, the drug that worked stops working. Which one do you try next?

Targeted cancer drugs fail by resistance — the tumor changes shape and the drug no longer binds. Deciding what to test next, or whether an already-approved drug could be repurposed, means expensive, slow wet-lab screening. This tool is the fast first pass: for a tumor's driver mutation it scores a panel of approved drugs against the wild-type and mutant protein and sorts them into what to stop, what still binds, and what's worth testing — with the uncertainty stated, before you spend a day at the bench.

Grounded in physicsEach drug is docked against the real mutant structure and rescored for binding affinity — not a guess from a text model.
Honest about uncertaintyEvery call carries a 95% credible interval from replicate runs. Affinity is a proxy, not a Kd; repurposing hits are hypotheses, never discoveries.
Reviewed by ClaudeDocking scores drugs that don't really bind. Claude weighs each hit by mechanism and flags the likely artifacts — so you test signal, not noise.
… or load a real, de-identified TCGA patient tumor

The verdict

Does it hold up?

The evidence — wild-type → mutant binding shift

Weakened Robust Improved Non-binder within ±1 kcal/mol — negligible ← binds better · 0 = no change · worse (resistance) →

Claude's read

Run it on your own tumor

Bring your own tumor — and your own GPU

You are not limited to these examples. The docking is the only GPU-heavy step, and it runs on your hardware (a lab cluster, or a rented cloud GPU). Give the tool your tumor's wild-type and mutant structures, run the documented pipeline, add one config entry, and the same triage and this page work on your genotype.

1
StructuresGet the wild-type and mutant PDB structures for your target.
2
Dock on your GPURun DiffDock + gnina from cluster/ against the drug panel.
3
Register itAdd one entry to config/mutations.json mapping the label to the structures.
4
TriageRebuild — your genotype now appears here with full credible intervals.
// config/mutations.json — add your genotype, no code changes
"EGFR L747_P753delinsS": {
  "target": "EGFR", "wt": "3POZ", "mut": "YOUR_PDB",
  "note": "exon-19 deletion; sensitizing"
}

Full walkthrough: docs/bring_your_own_tumor.md. The GPU run stays on your side — the tool ingests the scores it produces.