AI Digital Twins Predict Glioma Metabolism & Therapy Response | UMich Study
- A new study from the University of Michigan is leveraging the power of artificial intelligence to create “digital twins” of gliomas – aggressive and complex brain tumors –...
- Gliomas, which arise from the glial tissue of the nervous system, are notoriously difficult to treat.
- The research focuses on metabolic dependency, a growing area of cancer research that examines how tumors rewire their metabolic processes to produce energy and build new biomass.
Digital Twins Offer New Insights into Aggressive Brain Tumors
A new study from the University of Michigan is leveraging the power of artificial intelligence to create “digital twins” of gliomas – aggressive and complex brain tumors – offering a novel approach to understanding their metabolic pathways and potentially predicting treatment outcomes. Published , the research utilizes machine learning to construct a virtual replica of a patient’s tumor, allowing researchers to medically manipulate and study it in a way previously impossible.
Gliomas, which arise from the glial tissue of the nervous system, are notoriously difficult to treat. “A digital twin is a virtual representation of some physical property that exists in the real world,” explains Dr. Daniel Wahl, associate professor of radiation oncology and a co-author of the study. “Having a virtual representation of the system allows you to study and perturb it virtually, so you can determine what will happen to your real-life system if you make a modification.” This capability is crucial for a disease where treatment options are limited and often ineffective.
Understanding Metabolic Dependencies
The research focuses on metabolic dependency, a growing area of cancer research that examines how tumors rewire their metabolic processes to produce energy and build new biomass. These dependencies – influenced by genetic, environmental, and molecular factors – represent potential vulnerabilities that can be exploited therapeutically. The University of Michigan team specifically investigated two key pathways: nucleotide synthesis and serine consumption.
Nucleotide synthesis is the process by which tumors create the building blocks of DNA and RNA, essential for rapid proliferation. Serine, an amino acid, also plays a critical role in tumor growth and development. “The machine learning model could distinguish between these pathways and find the contribution for each of these pathways in each patient,” says Baharan Meghdadi, a doctoral student in chemical engineering and study co-author. “And so, based on the prediction of the model… we used a serine, glycine-free diet in those mice, and that helped to reduce the tumor size in those mice, and that experiment validates the prediction of our model.”
How Digital Twins are Built
The creation of these digital twins relies on a “first principles model” grounded in fundamental biochemical laws. Researchers developed stoichiometric and isotopic models – based on mass-balance constraints and known metabolic pathways – to simulate how a tumor processes nutrients. These models form the core framework of the digital twin. This framework is then combined with data from patients’ tumor cells to measure metabolic activity at the single-cell level, a process known as “flux” analysis. This allows physicians to gain a more detailed understanding of each tumor’s unique metabolic profile.
This detailed metabolic mapping is intended to address a significant challenge in cancer treatment: identifying which patients will respond to specific metabolic therapies. “We have no way to ask, really, ‘Is a metabolic pathway active in cancer?’” Dr. Wahl notes. “And so if we’re trying to develop metabolic therapies, we don’t know who to give (each) one to. And so that’s why we’ve developed this: now People can say, ‘Hey, patient one has this pathway active, we can block this pathway. Patient two doesn’t have that active? Well, we shouldn’t use that drug — we should do something else.’”
Bridging the Gap in Patient Data
The AI models used in the study are designed to overcome limitations in available patient data. “Our model uses the available data to develop this digital twin,” explains Dr. Deepak Nagrath, professor of biomedical engineering and another co-author. “If you want to directly use the limited patient data, we are unable to estimate the metabolism in vivo inside the patients. But using this machine learning-based approach, which is coupled with the first principles model — this hybrid approach allows us to expand and do the data importation and (increases) the span of the available patient data and address some of the issues.”
Next Steps: Clinical Trials and Expanded Patient Pools
The researchers successfully tested the serine-glycine metabolic model in mice trials, demonstrating that restricting serine intake could reduce tumor size in animals reliant on this pathway. The next step is to translate this technology into clinical trials, utilizing the AI-powered digital twin to guide treatment decisions in human patients.
However, the study was limited to a small cohort of eight patients. Dr. Nagrath emphasizes the need to expand the patient pool to validate the technology and make it more widely available to oncologists. “The data was only available to us for eight patients, and ideally we would want hundreds of patients,” he says. “But, as we grow and as there are resources available, we’ll be able to use it.”
This research represents a significant step forward in personalized cancer treatment, offering the potential to tailor therapies based on the unique metabolic characteristics of each patient’s tumor. By harnessing the power of AI and digital twin technology, researchers are paving the way for more effective and targeted approaches to combating this devastating disease.