Quantum mechanical approach to artificial intelligence could improve cancer outcomes

For children diagnosed with neuroblastoma, the most common infant cancer that occurs when early nerve cells grow out of control, the path to treatment is not easy. Some types of neuroblastoma may heal on their own, while others require aggressive intervention. Researchers have tried to tailor treatments to patients based on a single gene mutation, but with limited success. This is because patient outcomes depend on the entire molecular background, which includes millions or even billions of features, such as DNA and RNA from tissues and blood.

“It’s not just one gene; everything that’s happening inside a patient’s cells is important,” said Ory Alter, associate professor of biomedical engineering at the University of Utah’s Institute for Scientific Computing and Imaging.

Current artificial intelligence and machine learning (AI/ML) approaches require large amounts of training data, especially patient samples that are much larger than genetic features. For this reason, most clinical trials, which typically enroll only 20 to 100 people, are not very good at predicting patient outcomes. For example, a recent large-scale language model of the 30,000-nucleotide genome of the coronavirus required approximately 110 million samples. Translating this into the 3 billion nucleotide human genome, traditional AI approaches would require 33 trillion patients.

By using the mathematics of quantum mechanics, Alter and his collaborators have developed new AI/ML techniques that can improve treatment selection and drug success rates. their works are published in magazines Applied Physics Letters (APL) Quantum.

Features of billions of molecules

“Our quantum approach allows us to find relevant information from all layers of data, for example in the patient’s blood as well as the tumor,” Alter said. “We can take in and understand all of the millions to billions of molecular signatures of even a small number of patients. Therefore, we can understand disease mechanisms and predict drug targets to improve patient outcomes. We are also experimentally validating AI/ML predictions of targets and outcomes, widely considered the holy grail of biotechnology.”

The technique deploys a set of algorithms called multitensor comparative spectral decomposition, which Alter built on the quantum mechanical concepts of entanglement and superposition. Like a prism that splits white light into individual colors, this approach breaks down multiple layers of a patient’s molecular data, including their tumor, blood genome, and tumor (or the RNA messages that drive cancer growth) into linked patterns that predict health outcomes.

Alter and her team demonstrated their technology by analyzing open-source data from neuroblastoma cases. The algorithm discovered two new predictors of a patient’s life expectancy in response to treatment. These predictors consistently outperformed standard biomarkers across tumors, blood DNA, and tumor RNA. These findings hold true across separate groups of children treated at different times and hospitals, meaning the method can be applied to the general population to provide a clearer roadmap for patient care and drug development.

Developing more targeted treatments

“While neural network models are black boxes, our predictors are interpretable. They point to disease mechanisms and suggest genes to target to increase tumor treatment sensitivity,” Alter said. Her team also utilized the gene editing tool CRISPR-Cas9 to experimentally validate the prediction of outcome and drug targets for adult glioblastoma patients in clinical trials and preclinical studies.

Alter, an expert in computational medicine, holds an adjunct position in the US Department of Human Genetics and is a member of the Cancer Control and Population Sciences Research Program at the Huntsman Cancer Institute.

Her university-independent company, Prism AI Therapeutics, Inc., helps biotech and pharmaceutical companies better develop drugs by using algorithms and predictors to identify which patients will benefit most from clinical trials and which genes can be targeted to further improve outcomes.

Looking to the future, Alter hopes that as the team continues this research, it will be possible to apply it to individual patients. “This is the ultimate in precision medicine,” she said. “You have one person. Can you take data from that one person and come up with a treatment for that person? I think we can accomplish that.”

Alter is looking forward to other challenges as well. “Algorithms are completely data agnostic and could have endless applications beyond medicine,” she said, highlighting sustainable energy as one possibility.

The study was published June 10 in the journal Quantum Mechanics-Based Multitensor AI/ML Can Uniquely Discover, Validate, and Interpret Predictors from Noisy High-Dimensional Multiomic Data in Small Cohorts. applied physics letters (APL) quantum. Co-authors include Elizabeth Newman (Tufts University), Sri Priya Ponnapalli (Scale AI, Inc.), and Jessica W. Tsai (Children’s Hospital Los Angeles and Keck School of Medicine at the University of Southern California).

This research was supported by the National Institutes of Health (NIH) and the National Cancer Institute’s Oncology Physical Sciences Project “Multitensor Decomposition for Personalized Cancer Diagnosis and Prognosis,” the National Science Foundation (NSF) and the American Mathematical Institute Quantum Research Community Project “AIM Q,” the Musella Foundation in partnership with StacheStrong, the NIH Office of the Director, the NSF Division of Mathematical Sciences, and Alex’s. The Lemonade Stand Foundation, Rally Foundation, and St. Baldrick’s Foundation partner with Griffin’s Guardians.