Moth Hack 2026 · Quantum Chronicle · 2026-10-06

We put a protein on a quantum computer

KRAS was undruggable for 40 years — until a drug found a hidden pocket in its shape. We gave that shape to a real IBM quantum chip and measured what survived.

immutable copy · lens://d2cc8601c0b3c728815ccb15e87a5b18351e0063574d417acffa39e33eb83d34

In plain words

The protein KRAS drives roughly a quarter of human cancers and resisted drugs for four decades — until sotorasib found a hidden pocket written into its shape, not its active site. That shape is contact topology: which residues touch which. We distilled the allosteric cores of KRAS G12C and cardiac myosin into 20-qubit coupling maps and ran them on Moth's graph-v1 engine — once on an emulator, once on IBM's ibm_fez quantum processor. The result was a clean split: correlation magnitude survives real hardware (about 84% of emulator strength) while direction decoheres to essentially zero, and the protein-versus-random-control separation washes out. Fine structure is NISQ noise's first casualty. We also sonified both measurements — each coupling edge becomes a note, so you can literally hear the decoherence — and every claim on the page links to a raw job receipt.

New words

allostery — Control at a distance — a hidden pocket changes shape and the protein's active site answers.
contact topology — Which residues physically touch which — the protein's wiring diagram.
decoherence — Noise washing out a quantum state — what real hardware does to ideal answers.

Where this comes from

Interactive exhibit — /quantum/ on the Observatory quantum

Kytos Observatory

c008 experiment record — receipts, reproduce scripts, figures c008-moth-qpu-tomography

Kytos / Cleveland GQAI workstream

Moth Atlas — engine platform used for all measurements platform.mothquantum.com

Moth Quantum

Deep science

KRAS G12C and cardiac myosin contact graphs were coarse-grained to their 20-node allosteric cores (catalytic site, known drug pocket, strongest couplings; 50 and 42 edges) plus a degree-matched random control. Submitted to Moth graph-v1 in emu and qpu modes on IBM ibm_fez: 1024 shots, full two-qubit tomography per edge. Measured: mean edge |corr| held ~0.84× of emulator on hardware while Bloch-vector direction cosine fell to ≈0; emu separated protein graphs from the random control, qpu did not — topology-specific fine structure is the first casualty of NISQ noise. Read-back: writing measured correlations through qpixl-v1 returned Pearson 0.98. Sonification maps each edge to a note in submission order; emu and qpu render the same graph as audibly different melodies. A custom Laplacian CTQW was compiled to OpenQASM 2 and verified vs exact expm at Pearson 0.991 — execution blocked only by a tomography-api-v2 outage during the hack window. Engines: graph-v1, qpixl-v1, telablur-v1, retrocausal-echo-v1.