Researchers from Cleveland Clinic and IBM have jointly developed a framework for using quantum computing to make predictions of antigen presentation and immunotherapy response, according to the study results published in Science Advances.
The framework, called Quantum Convolutional HLA Immunogenic Peptide Prediction (Q-CHIPP), demonstrates both the feasibility and promise of a quantum machine learning approach to biomedical and immuno-oncology research.
“The creation of this model exemplifies team science,” said senior author Tyler J. Alban, PhD, Center for Immunotherapy and Precision Immuno-Oncology, Cleveland Clinic. “This was a multidisciplinary team where we had experts in cancer, computational sciences and quantum computing, all at the table teaching each other and learning from one another. We could not have achieved this without a team approach.”
"This collaborative approach exemplifies how Cleveland Clinic and IBM are pioneering quantum computing in immunology and broadly speaking in healthcare,” said corresponding author Sara Capponi, PhD, Senior Research Scientist, IBM Research. “Only a team combining this breadth of expertise could translate the immune system's features into quantum circuits, ensure the underlying math faithfully captured the biology, and validate the results against real-world patient data.”
Background and Study Findings
Quantum computing has been of interest for solving complex biological problems, such as drug discovery, protein folding, etc. Researchers applied quantum computing to neoantigen prediction and immunotherapy response to advance immuno-oncology research.
To address the limitations of noisy datasets in immunology that classical computers could not solve, the researchers applied noise mitigation techniques and controlled shot-based sampling to train the data on real hardware and in a warm start hybrid approach.
They used quantum convolutional neural networks for MHC binding and immunogenicity prediction to assess classification accuracy in comparison with classical computer approaches. This led to the development of Q-CHIPP for integrating MHC binding and T-cell recognition, which targets HLA-A*02:01–restricted 9-mer peptides. The framework was also able to identify which peptides were immunogenic.
The model is able to learn from training sets as small as only ~150 samples, as with Cleveland Clinic's atlas of neoantigens possibly able to trigger immune responses. The researchers were also able to scale their framework to full-length peptide modeling using only 46 qubits of quantum hardware.
Additionally, the model outperformed traditional computing methods, showing a 6% increase in accuracy, in tests of similar constraints and parameter lengths.
Going forward, the research team is planning to further enhance their model to more accurately identify therapeutic targets.
DISCLOSURES: For full disclosures of the study authors, visit science.org.
