A new machine learning-based approach developed at the University of Michigan to map tumor metabolism in brain tumor patients in real time could help doctors discover which treatment strategies are most effective for individual cases of glioma. The team validated the model’s accuracy by comparing it to human patient data and performing mouse experiments.
The study, published in the journal Cell Metabolism, builds on previous research showing that a patient’s diet can slow the progression of some gliomas. Some tumors cannot grow if a patient does not receive certain protein building blocks called amino acids. However, other tumors can produce these amino acids themselves and can continue to grow anyway. Until now, there was no easy way to determine which patients would benefit from dietary restriction.
Because some cells can obtain their molecules from the environment, the digital twin’s ability to map metabolic activity within tumors also helped determine whether drugs that prevent tumors from producing the building blocks to replicate and repair DNA would work.
To overcome the challenges of mapping tumor metabolism in the brain, a team from Michigan has developed a computer-based “digital twin” that can predict how an individual patient’s brain tumor will respond to each treatment. These were primarily funded by the National Institutes of Health, specifically the National Cancer Institute.
“Typically, metabolic measurements during surgery to remove a tumor do not provide a clear picture of a tumor’s metabolism. Surgeons cannot observe how metabolism changes over time, and laboratories are limited to studying tissue after surgery. By integrating limited patient data with a model based on basic biology, chemistry, and physics, we overcame these obstacles,” said Deepak Nagrath, a UM biomedical engineering professor who co-authored the study.
Digital twins use patient data obtained through blood draws, metabolic measurements of tumor tissue, and genetic profiles of tumors. The digital twin then calculates the rate at which cancer cells consume and process nutrients, known as metabolic flux.
“This is the first time that machine learning and AI-based approaches have been used to directly measure metabolic flux in a patient’s tumor,” said Baharan Meghdadi, a doctoral student in chemical engineering and co-lead author of the study.
The researchers built a type of deep learning model called a convolutional neural network and trained it on synthetic patient data generated based on known biology and chemistry and constrained by measurements from eight glioma patients who were injected with labeled glucose during surgery. By comparing their computer model with different data from six of those patients, they found that the digital twin could predict metabolic activity with high accuracy. In experiments conducted on mice, the researchers confirmed that the diet was sufficient to slow tumor growth in the mice that the digital twin had identified as suitable candidates for treatment.
“These results are exciting. Being able to measure the metabolic activity of a patient’s tumor may allow us to predict which metabolic therapy will be most effective for each patient,” said Daniel Wahl, Achtenberg Family Professor of Radiation Oncology and co-author of the study.
The digital twin also predicted how tumors would respond to mycophenolate mofetil, a drug that targets the way cancer cells build their DNA. The digital twin pinpointed that some tumors can evade the effects of drugs by taking in nutrients from their surroundings using “salvage pathways.” Again, the research team confirmed their predictions in mouse experiments.
“This wonderful tool will help physicians avoid prescribing treatments to which certain tumors are already resistant, and will provide a path toward more targeted and personalized treatments for patients,” said Wajid N. Al-Khorou, assistant professor of neurosurgery and co-lead author of the study.
Doctors can use a patient’s digital twin to test whether a particular diet or drug actually starves the cancer before the patient changes their diet plan or takes a new drug.
“This study brings us closer to truly personalized cancer treatments, not just for brain tumors, but ultimately for a variety of tumors. By simulating different treatments virtually, we hope to save patients from unnecessary treatments and allow them to focus on treatments that are likely to be effective,” said Kostas Lysiotis, professor of oncology at the Maisel Institute and co-author of the study.
reference: Scott AJ, Mittal A, Meghdadi B Rewiring of cortical glucose metabolism promotes human brain tumor growth. nature. 2025;646(8084):413-422. doi: 10.1038/s41586-025-09460-7
This article has been reprinted from the following material: Note: Materials may be edited for length and content. Please contact the citation source for details. You can access our press release publishing policy here.
