Researchers at the University of Utah have developed a quantum-inspired AI approach that could significantly improve precision oncology by tailoring treatments to a patient's complete molecular profile rather than relying on single genetic mutations. The technique enables researchers to analyze millions to billions of molecular features from only a small number of patients, overcoming one of the biggest limitations of conventional AI in clinical research.
The proof-of-concept study focused on neuroblastoma, the most common solid cancer in infants. According to the researchers, the new method could not only improve treatment selection for individual patients but also accelerate the discovery of new drug targets and increase the success rate of clinical trials.
Looking beyond single-gene biomarkers
Precision medicine has traditionally focused on identifying individual genetic mutations that can guide treatment decisions. However, for many cancers, including neuroblastoma, clinical outcomes are determined by far more than a single gene. "It's much more than just one gene. Everything that's happening in the cells of the patient matters," said Orly Alter, associate professor of biomedical engineering at the University of Utah's Scientific Computing & Imaging Institute.
A patient's disease biology consists of multiple interconnected layers of information, including DNA, RNA and molecular signals from both tumour tissue and blood. Together, these datasets contain millions or even billions of biological features. Conventional machine learning methods struggle with such complexity because they typically require vastly more patient samples than molecular variables. While that may be feasible for internet-scale applications, it is unrealistic for most medical studies, where clinical trials often include only 20 to 100 participants.
The researchers illustrate this limitation by comparing their work with a large language model trained on the 30,000-nucleotide genome of the COVID-19 virus, which required approximately 110 million samples. Scaling such an approach to the three-billion-nucleotide human genome would theoretically require trillions of patient datasets.
Quantum mathematics
Instead of relying on ever-larger datasets, the research team applied mathematical principles inspired by quantum mechanics, including entanglement and superposition, to create a new AI framework. The technique, known as multitensor comparative spectral decomposition, simultaneously analyzes multiple layers of molecular information from each patient. These include tumour DNA, blood DNA and tumour RNA, allowing the model to identify linked biological patterns associated with treatment response.
"Our quantum approach allows us to find the relevant information in every layer of the data, for example, from the patients' blood in addition to their tumors," Alter explained. "Even for very few patients, we can still take everything in, their millions to billions of molecular features, and make sense of them. We can, therefore, understand the disease mechanisms and predict drug targets to improve patients' outcomes. We also validate our AI/ML predictions of targets and outcomes experimentally, which is widely considered a biotechnology holy grail." Unlike many deep learning models, which often function as "black boxes," the new approach produces interpretable results that can be linked directly to underlying disease mechanisms.
New biomarkers and drug targets
The researchers evaluated the method using publicly available neuroblastoma datasets. Their algorithms identified two previously unknown predictors of patient survival that consistently outperformed established biomarkers across tumour DNA, blood DNA and tumour RNA. Importantly, these findings were reproduced in independent groups of children treated at different hospitals and during different time periods, suggesting that the approach is robust across patient populations rather than being limited to a single dataset.
The team has also applied the technology to adult glioblastoma, where predicted treatment targets and patient outcomes were experimentally validated using CRISPR-Cas9 gene-editing techniques in both preclinical studies and clinical trial settings. According to Alter, this combination of computational prediction and biological validation distinguishes the platform from many existing AI approaches in precision medicine.
Truly personalized oncology
Beyond identifying new biomarkers, the researchers believe the technology could improve drug development by helping pharmaceutical companies identify patients most likely to benefit from specific therapies and by revealing additional molecular targets that could enhance treatment effectiveness. The algorithms are already being translated through Prism AI Therapeutics, a University of Utah spin-off company that applies the technology to support biotechnology and pharmaceutical research.
Ultimately, the researchers envision extending the approach to the individual patient level, where treatment decisions could be based on the complete molecular profile of a single person rather than population averages. "That's the ultimate precision medicine," Alter said. "You have a single person. Can you take the data from just that one person and come up with a treatment for them? I think we can get there."
Although the current work focused on cancer, the researchers note that the algorithms are data-agnostic and could eventually be applied to many other complex scientific challenges beyond medicine, including fields such as sustainable energy.
AI for personalized cancer treatment
In 2025 researchers from the Icahn School of Medicine at Mount Sinai, Memorial Sloan Kettering Cancer Center and European partners demonstrated that AI can accelerate lung cancer diagnostics by predicting EGFR mutations directly from routine H&E-stained biopsy slides. Tthe study used the largest pathology dataset of lung adenocarcinoma cases combined with genomic data from multiple centers in the U.S. and Europe.
In a real-world “silent” clinical trial, the AI model accurately identified EGFR mutations without influencing clinical decisions and showed the potential to reduce urgent genetic testing by more than 40%. According to the researchers, this approach could speed treatment decisions, lower costs and improve access to targeted therapies, particularly in hospitals with limited molecular testing capabilities. Future work will validate additional biomarkers, expand the technology to other tumor types and evaluate its use in resource-limited healthcare settings.
