DANISH RESEARCHERS USE QUANTUM COMPUTING AND AI TO DESIGN PROTEIN FRAGMENTS FOR FUTURE CANCER VACCINES
A team at the Technical University of Denmark, working with ORCA Computing and Sparrow Quantum, has used a quantum computer to help an AI model design small protein fragments that the human immune system can recognise. They then took the AI’s designs into a lab and confirmed many of them worked. Corresponding author Timothy Patrick Jenkins told The Indian Express that the technology is still far from patient use, but could eventually help build more personalised cancer vaccines. The study is a preprint on bioRxiv, awaiting peer review.
To follow what the team did, it helps to know what a cancer vaccine actually asks the body to do. Every cell in the body puts up little markers on its surface, like ID cards. Healthy cells show normal cards. Cancer cells often show unusual ones, and the immune system, when trained to recognise them, can attack. A personalised cancer vaccine works by teaching one patient’s immune system what their tumour’s specific ID card looks like. Designing that card, a short protein fragment called a peptide, is the first step. The problem is that there are hundreds of billions of possible fragments, and only a small share will actually fit on a person’s cells and be seen by the immune system.
AI helps by proposing new fragments based on ones that already work. Every AI model needs a random starting input to begin its search, like a chef starting with a mystery basket of ingredients. The Danish team replaced the usual random input with patterns generated by a photonic quantum computer, which uses light particles instead of ordinary chips. The quantum machine did not design the fragments. It gave the AI a different, more structured starting basket to work from.
Trained on 106,000 known pairings, the model was asked to generate 1,000 candidates for each of 131 immune-system variants. The quantum-assisted version outperformed the standard approach on 63% of variants. The biggest gains showed up for rare immune types that are poorly represented in existing data, the group that current AI tools have consistently served least well. For one such variant, the quantum-based model produced roughly twice as many likely binders. When 20 top peptides for three rare variants were tested in the lab, all 20 held stable complexes for two of them. The third variant, known to be difficult to predict, showed mixed results.
The team is explicit that this is not proof of quantum advantage. The problem was small enough that regular computers could still simulate it, and stable binding is only the first step toward a working vaccine. Still, the direction matters for personalised cancer therapy. A vaccine designed around common immune profiles may not work well for people whose immune systems carry less common variants. Quantum-generated randomness, plugged into an existing AI, helped the model perform best in exactly the corner where personalised vaccine design has been weakest.
Source: The Indian Express, Quantum Insider, bioRxiv preprint
