Timothy Patrick Jenkins and ORCA Computing CEO Richard Murray have something the quantum-computing market badly needs: a wet-lab result tied to a near-term use case. A Technical University of Denmark team led by Jenkins used a hybrid generative-AI and quantum-computing workflow to design novel MHC class I-binding peptides, then validated selected designs in vitro, according to a WIRED story and a bioRxiv preprint posted July 9.
Aligned News thread on X The claim is narrower than the quantum sales pitch usually sounds, which is why it is useful. The researchers did not discover a drug. They did not show a quantum computer replacing classical drug-discovery infrastructure. They showed that a generative adversarial network seeded with samples from a real photonic quantum processor could produce peptide candidates that outperformed classical baselines on a specific immune-binding task, especially for HLA alleles with scarce training data. The preprint says the team tested the approach across many HLA alleles in silico and then selected understudied alleles for laboratory validation using peptide-MHC stability assays.
Jenkins is an associate professor in DTU's Department of Biotechnology and Biomedicine, where his research profile spans antibody technologies, digital biotechnology and biologics engineering. WIRED reports that his group uses big data and AI to discover proteins that could make immunotherapies cheaper and faster, with a particular problem: the medical data underlying many biological models is skewed toward Western populations. That makes the result strategically important for Jenkins because low-data targets are exactly where classical models tend to struggle and where overlooked patient groups can disappear from the training set.
WIRED reported that Jenkins' team worked weekends and pooled unspent money from other projects to run the experiment. Jenkins told WIRED he had been a quantum skeptic and thought practical applications for his work were decades away. The turn came after his team learned that quantum systems had improved diversity in generative image models and asked whether the same idea could help peptide generation.
The actual design loop
The workflow used a generative adversarial network, or GAN, for de novo peptide design. In a standard classical setup, the model samples from a conventional prior distribution. In the DTU-ORCA setup, the model used quantum-derived latent vectors from a photonic quantum processor. The preprint describes the work as the first end-to-end hybrid quantum-classical pipeline for de novo design of MHC class I-binding peptides.
The core finding is about exploration. The preprint says quantum-derived priors increased the yield of predicted strong binders, with the largest relative gains for understudied alleles where classical baselines performed worst. On alleles chosen for further evaluation, the researchers confirmed in vitro that quantum-designed peptides stabilized peptide-MHC class I complexes.
ORCA has also published the Pep-Q-GAN GitHub repository, an Apache-2.0 implementation of the paper's classical and hybrid quantum-classical GANs. The repo gives outside researchers a concrete artifact to inspect: training scripts, inference scripts, NetMHCpan binding-analysis hooks and configuration options for quantum latent distributions. The README says quantum samples used for training can be obtained by emailing ORCA, so the published code does not make the hardware portion fully self-serve.
What ORCA gets out of it
For ORCA Computing, the DTU result gives Murray a sharper answer to the most common buyer question in quantum: what can the hardware do before fault tolerance arrives? ORCA, a London-based University of Oxford spinout founded in 2019, builds photonic quantum systems. Its site says the PT Series is designed as a rack-mounted, room-temperature machine that can sit inside existing high-performance-computing infrastructure, with a Python SDK integrated with PyTorch. That infrastructure positioning matters. ORCA is not pitching drug companies a standalone black box that replaces their discovery stack. The peptide work is framed as a hybrid workflow: classical models generate and score, quantum-derived sampling changes the search behavior, and lab assays decide whether any of it matters. That is closer to how R&D teams actually buy tools. A small biological-design team does not need a grand proof of quantum advantage to care about a workflow that produces a better shortlist for targets where the data is thin.
Murray told WIRED that industrial companies often see quantum as hazy and far away because the field has lacked clear near-term examples of usefulness. He framed the DTU work as one such example. ORCA is also applying its technology in projects with BP on chemistry and Toyota on design-process efficiency, according to WIRED.
The limits are the point
The result also keeps the caveats in view. WIRED reported that current quantum computers are still too small to run full-scale, cutting-edge AI models, and that better results could currently be achieved on a classical computer for larger model complexity. DTU PhD student Jonathan Funk told WIRED the quantum computer could not encode the complexity of a normal-sized antibody, which is what the team usually works with.
That limitation explains why the experiment focused on peptides. Peptides are shorter chains of amino acids, and binding to a target protein is only one step in vaccine or immunotherapy development. A peptide that stabilizes an MHC complex still has to move through a much longer chain of biological evidence before anyone can talk about efficacy, dosing, safety or clinical use.
The bioRxiv paper is also a preprint. Its claims have not gone through journal peer review. The author list includes researchers from DTU, ORCA Computing, Sparrow Quantum, the Poznan Supercomputing and Networking Center and other institutions. The competing-interest statement says several authors were employed by ORCA Computing or Sparrow Quantum while preparing the manuscript, a relevant disclosure given the commercial implications of the result.
The stronger read is that Jenkins' group and ORCA have produced an early, testable design pattern rather than a finished drug-discovery platform. The pattern is clear: use AI to propose candidates, use quantum-derived sampling to push the model into less obvious regions of sequence space, then spend wet-lab money only on the shortlist. If that pattern holds up across larger proteins and newer models, it could matter most in the parts of medicine that lack both rich data and rich funding.