A quantum mechanics approach to artificial intelligence can improve cancer outcomes – @theU
Orly Alter of the University of Utah developed a quantum mechanics-based AI/ML technique that predicts cancer outcomes and drug targets using small patient cohorts. Published in Applied Physics Letters (APL) Quantum, the method uses multitensor comparative spectral decompositions to analyze billions of molecular features, bypassing the massive data requirements of traditional AI.
Why does traditional AI struggle with rare cancers?
Most AI and machine learning models need massive datasets to function. According to Orly Alter, an associate professor at the University of Utah’s Scientific Computing & Imaging Institute, traditional approaches require far more patient samples than genetic features. This creates a mathematical wall for clinical trials, which often only enroll 20 to 100 people.
Alter points to a stark contrast in data needs. A large language model for the COVID-19 virus genome required about 110 million samples for its 30,000-nucleotide genome. If that same logic applied to the 3-billion-nucleotide human genome, Alter calculates that a conventional AI would need 33 trillion patients to achieve similar results.
How does quantum-based AI analyze billions of features?
The new technique uses the concepts of entanglement and superposition from quantum mechanics. Alter describes the process as being similar to a prism splitting white light into colors. The algorithms, called multitensor comparative spectral decompositions, break down layers of molecular data—including blood and tumor genomes and RNA—into linked patterns.
This approach allows researchers to identify relevant information across every layer of data, even with very few patients. In a study using open-source neuroblastoma data, the team discovered two new predictors of life expectancy. These predictors outperformed standard biomarkers across tumor and blood DNA, and the results remained consistent across different hospitals and time periods, according to the APL Quantum report.
Comparing AI Architectures
| Feature | Traditional Neural Networks | Quantum-Based Multitensor AI |
|---|---|---|
| Data Requirement | Massive (millions of samples) | Small cohorts (20-100 patients) |
| Transparency | “Black box” (uninterpretable) | Interpretable (points to mechanisms) |
| Analysis Scope | Often limited to single-gene mutations | Billions of multiomic features |
What happens when AI becomes interpretable?
Unlike neural networks that provide an answer without explaining the “why,” Alter’s predictors are interpretable. They point directly to disease mechanisms and suggest specific genes to target to make tumors more sensitive to treatment.
The team didn’t just stop at predictions. They used CRISPR-Cas9 gene-editing tools to experimentally validate drug targets and patient outcomes for adult glioblastoma in clinical trials and preclinical studies. This ability to predict a target and then prove it in a lab is what Alter describes as a “biotechnology holy grail.”
Can this lead to “N-of-1” precision medicine?
The long-term goal for this research is the shift toward individual-level treatment. Alter believes the technology could eventually move from analyzing small groups to analyzing a single person. By taking the molecular data from one patient, doctors could potentially develop a custom treatment plan tailored to that individual’s specific genetic makeup.
This transition is being commercialized through Prism AI Therapeutics, Inc., a University of Utah spinoff. The company works with pharmaceutical firms to identify which patients are most likely to benefit from a clinical trial and which genes should be targeted to increase the success rate of new drugs.
Will these algorithms work outside of oncology?
Because the algorithms are “data agnostic,” they aren’t limited to cancer research. Alter suggests the mathematical framework could be applied to any field dealing with high-dimensional, noisy data. One primary area of interest she highlighted is sustainable energy, where similar complex patterns must be decoded to improve efficiency.
Frequently Asked Questions
What is the main advantage of quantum-based AI in medicine?
It allows for accurate predictions using small patient groups (cohorts) by analyzing billions of molecular features, whereas traditional AI requires millions of samples.
How was the neuroblastoma research validated?
The findings were validated using open-source data across multiple hospitals and through experimental testing with CRISPR-Cas9 gene editing.
What is a multitensor comparative spectral decomposition?
It is a set of algorithms based on quantum entanglement and superposition that breaks down complex, multi-layered data into linked patterns to predict health outcomes.
Do you think “N-of-1” medicine will become the standard of care in the next decade? Let us know your thoughts in the comments or subscribe to our newsletter for more updates on biotech breakthroughs.