A quantum framework for cancer immunotherapy
Cleveland Clinic and IBM researchers published results for Q-CHIPP, short for Quantum Convolutional HLA Immunogenic Peptide Prediction, a quantum machine learning approach designed to forecast which tumor mutations will trigger an immune response. The method folds antigen presentation and immunotherapy response prediction into a single model, and the team's paper appeared in Science Advances. The researchers reported that the quantum convolutional neural network outperformed classical computing methods when evaluated under similar constraints and parameter limits, and was able to extract meaningful biological patterns from training datasets as small as 150 samples — a useful property in immuno-oncology, where labeled patient data is often limited.
Scaling to biologically realistic peptides
The group scaled the approach to model full-length peptides on quantum hardware using 46 qubits, moving beyond reduced test systems toward biologically relevant inputs. Sara Capponi, a corresponding author and senior research scientist at IBM Research, said the collaboration translated features of the immune system into quantum circuits, verified the underlying math against the underlying biology, and validated results against real-world patient data. The team plans to continue refining Q-CHIPP to better identify therapeutic targets, with the stated aim of accelerating personalized cancer immunotherapies and next-generation cancer vaccines.
AI-designed circuits for pharmaceutical molecules
Separately, researchers from Quantinuum, NVIDIA and Pfizer described an artificial intelligence framework that generates quantum circuits for molecular simulations thousands of times faster than ADAPT-VQE, a leading quantum chemistry algorithm, while matching or exceeding its accuracy on benchmark tests. The framework combines transformer-based language models with reinforcement learning to produce complete molecular ground-state preparation circuits in a single inference step, replacing the iterative optimization used in conventional quantum chemistry workflows. The team selected imipramine, a tricyclic antidepressant with multiple reactive sites and many three-dimensional conformers, as its primary pharmaceutical test case.
Hardware runs and broader quantum advantage
Representative AI-generated circuits were executed on Quantinuum's Helios trapped-ion quantum computer, and the researchers reported that the reinforcement learning stage frequently produced circuits with lower calculated energies than those in the original training set, indicating the model had discovered improved solutions rather than simply reproducing ADAPT-VQE. Both medical-focused studies were published on July 30, 2026 alongside a broader quantum advantage claim from IBM and the University of Chicago, whose team encoded 70 logical qubits, ran 2,415 logical two-qubit operations and 468 logical T gates, and finished a verifiable computation in roughly 15 minutes that leading classical simulation methods could not match within feasible time. Taken together, the results illustrate quantum hardware and algorithms being aimed at concrete biomedical problems, though the medical work remains at the research rather than clinical stage.
Share this article
