A research team at the Technical University of Denmark has shown that a quantum computer can improve the accuracy and reach of generative artificial intelligence models used in drug discovery. The work focused on creating novel peptides, short chains of amino acids that can bind to specific proteins in the body.
The result is not a finished drug, and the researchers are clear that quantum computers remain limited. But the study matters because it connects a much-discussed technology to a practical biological workflow, then checks the output in the laboratory.
A hybrid route to peptide design
The DTU team ran a generative AI model for predicting proteins together with a printer-sized quantum computer built by British startup Orca Computing. The setup linked a quantum machine with traditional processors, creating a hybrid method rather than replacing conventional computing entirely.
That hybrid technique was used to generate new peptides capable of binding to specific proteins in the body. In vaccine development, finding molecules that can bind to relevant targets is an important step, even though it is only one part of a much larger process.
The key claim is not that quantum computing has suddenly solved drug discovery. It is narrower and more useful: in this experiment, the quantum-assisted model produced more successful peptides than the classical counterpart after the peptides were made and tested in the lab.
The strongest improvements appeared where training data was rare. That point is central because many biological AI systems are limited by the amount and representativeness of the data they can learn from.
Why scarce data is the hard case
Timothy Patrick Jenkins, the DTU professor who led the project, works with big data and AI to discover proteins that could unlock new immunotherapies cheaper and faster. His team is often funded by the Novo Nordisk Foundation.
Like many biological model builders, the team wants more data. But its challenge is more specific: medical research has focused heavily on Western populations, leaving gaps in data that represents the full variety of genetic information across the human race.
That gap can make it harder to develop peptides that work for understudied populations, including those in Asia and Africa, according to Timothy Patrick Jenkins. In that context, a model that can generate a more diverse set of peptide candidates could be valuable.
The researchers formed their hypothesis after learning that quantum machines had produced a similar effect in image generation. They wanted to see whether embedding a quantum computer into their workflow could help the model explore more possibilities, especially when the target had less data behind it.
- Problem: biological AI models often need more and broader training data.
- Test: combine a generative AI protein model with a quantum computer.
- Output: novel peptides designed to bind to specific proteins.
- Result: lab testing showed more successful peptides than the classical model, with the clearest gains where data was rare.
Why the team had to prove it in the lab
Quantum computing is still a nascent field, and it faces heavy scrutiny. The machines are technically difficult to build, and applying them to real problems has remained a major challenge.
Timothy Patrick Jenkins was not always convinced the technology could help his own work. He said, “I was a huge quantum skeptic” and believed any useful application would be “decades away.”
That skepticism shaped the project. The team did not stop at model output. It made the peptides in the laboratory and tested whether they would bind to the particular proteins.
That real-world validation was essential because a prediction is not the same thing as a working biological result. As Timothy Patrick Jenkins told WIRED, “We needed to really prove it to convince skeptics that our predictions connect to the real world.”
The way the project came together also says something about how risky early-stage science can be. The researchers worked weekends and pooled unspent money from other projects because, in Timothy Patrick Jenkins’ words, “most innovative science is too scary for foundations.”
The limits are still large
The findings do not mean quantum computers are ready to run the largest and most advanced AI drug discovery systems. The source makes clear that the new process will not revolutionize research yet because quantum computers are still too small to run full-scale, cutting-edge AI models.
That means better results could still be achieved on a classical computer. The current value of the work is as a proof that the hybrid workflow can produce measurable biological outputs, not as proof that quantum systems are already superior across the field.
DTU PhD student Jonathan Funk put the limitation plainly: “Quantum is still not very powerful, so the level of complexity that we could encode wasn’t a normal-sized antibody, which is what we usually work with.”
There is another practical limit. Finding a peptide that can bind to a specific gene is only one step in vaccine development. It would not, by itself, create a successful drug.
Those caveats matter because they keep the result in proportion. The work is promising because it shows an approach that may improve generative AI in difficult data conditions. It is not a shortcut around the long process of developing and validating medicines.
What comes next for quantum-assisted AI
Orca Computing chief executive officer Richard Murray told WIRED that many industrial companies see quantum as “hazy and far away,” partly because the technology “has not ever had really clear near-term examples of usefulness.”
He argues that this study is novel because it points to a near-term commercial application for quantum computing. Orca Computing is also applying the technology through projects with oil major BP on chemistry and carmaker Toyota on making its design process more efficient.
The DTU team now plans to test whether the workflow can be used with more cutting-edge models and larger proteins. Timothy Patrick Jenkins described the study as an easier way to validate whether the team has “a shot at moving the needle substantially.”
He also said generative AI workflows are particularly valuable in neglected diseases that receive little research money. In addition, he is looking at using a quantum computer to enhance his generative AI method for designing synthetic antidotes for snakebite venom.
The larger implication is careful but important. If quantum-assisted generative AI can help researchers generate useful candidates when data is thin, it could become a practical tool in areas where conventional biomedical datasets are weakest. For now, the case is strongest as an experimentally tested step forward, not as a finished transformation of drug discovery.