Drug discovery you can open.
Fold the target, generate candidates, predict affinity, search conformers, then build the clinical evidence through to an IND package — nine steps, each one sealed at the model layer as it runs. Your team reviews the record before anyone touches a bench, and a third party can check it without taking our word for it.
How it works
Nine steps. Sealed as they run.
Two chains: discovery, from the target to a ranked molecule, and the clinical evidence built on it, from the safety signal to an IND package. Nothing here runs a model for a visitor. Every panel shows what a real run produced, and the seal is written while the step executes — not reconstructed afterwards.
Pharma · discovery
Fold the target
Build a 3D structure of the protein you care about, so the later steps work from a real shape rather than a guess.
Boltz-2 (MIT) · EpochCore serversGenerate candidates
Propose small molecules aimed at that fold. You see what was suggested and how it sits in the pocket.
GenMol (NVIDIA) · EpochCore serversPredict affinity
Estimate how well a surviving candidate may bind. Predicted — not a wet-lab measurement.
Boltz-2 co-fold (MIT) · EpochCore serversSearch conformers
Search the winning molecule’s low-energy shapes. The calibrated router picks the machine and the sealed record names which one ran it. A simulation today, not a quantum-advantage claim.
QAOA, simulated · EpochCore routerRx · clinical evidence
Safety signal
Pull adverse-event signals from openFDA FAERS, so safety context sits on the same trail as the rest of the evidence.
openFDA FAERSEvidence base
Assemble the clinical literature and prior findings your team will argue from — inspectable, not a black box.
ClinicalTrials.gov · PubMedTrial design
Size the study, prioritise the arms, and lock the design assumptions before enrollment spend.
QAOA · IBM HeronIND package
Lay the sealed record against the IND sections of 21 CFR 312.23, so your team sees what is in hand and what still needs the sponsor’s own data. A manifest, not a filing.
21 CFR 312.23WORM clinical seal
Write the clinical record to a tamper-evident, write-once seal, so compliance can open the trail later without taking our word for it.
Write-once by design, uncertifiedModel layer · on by default
Sealed at the model. Built for audit.
Each step writes a signed, tamper-evident record as it executes: a digest over the step’s own output, three signatures over it — Ed25519, and the post-quantum ML-DSA-87 (FIPS 204) and SLH-DSA-128f (FIPS 205) — and a link to the step before it. The archive is write-once — a published run cannot be edited later, including by us.
Every day, one fingerprint of all of that day’s seals is timestamped in Bitcoin through OpenTimestamps and anchored on Base, so the date of a record does not rest on our word either. Only the fingerprint goes on a chain; results and molecules never do.
A reader can fetch the chain, re-verify every signature against our published keys in their own browser, and confirm the links are intact. Compliance framing is design, uncertified — we are not attesting to a standard on your behalf.
Qubit-aware routing
Time and money on the right machine.
Most work belongs on GPU, and that is where it stays: the discovery chain’s fold, generation and co-fold run on GPUs on EpochCore’s own servers, and its conformer search runs as a simulation with the router’s verdict sealed beside it. When work does earn it, it goes to a real QPU. The clinical chain’s trial-design step runs QAOA on IBM Heron — each run takes the least-queued device from the live queue at submit, and seals the job id, the device and the queue it chose from with the result. The pharma build’s own hardware validation ran on ibm_kingston. All of it is in the sealed records.
Behind that, and the corpus the routing weights were fitted on: 1,605 real IBM Heron jobs run between 9 February and 2 July 2026, every one counted, failures included, with circuits up to 120 qubits wide that ran to completion — summarised and fingerprinted, with the full per-job manifest available under NDA, and a plain statement of what it proves and what it doesn’t.
Every circuit in it is pre-compiled — transpiled to the device’s native gate set before it runs — and the depths and gate counts are published with it. That is the part you can check against IBM’s own device specifications without an account.
Two ways in
Discovery or trial.
Kill the bad molecule early
Screen out non-starters before they burn years and budget. Fold, generate, pre-screen for drug-likeness, then co-fold the survivors.
Open Quantum Pharma →Carry the dossier
Assemble the safety signal and the evidence base your team will argue from, and keep the trail intact and inspectable.
Open Quantum Rx →Who it’s for
R&D labs. Built to be checked.
- Discovery and computational chemistry groups evaluating new targets
- Biotech R&D leads who want a fold-to-binding trail before wet-lab spend
- Partner teams that need a shared, inspectable run — not a black box
Every affinity number on this site is a prediction from a model, not a wet-lab measurement. This is a research tool for discovery teams. Not for clinical care or patient decisions.
What you can do today
Look first. Or send a target.
Open the live viewer and walk a saved fold, candidate and binding step in your browser — no signup. Or tell us the protein you care about and we will hand-run a discovery pass and return a link your team can open.