A team of researchers from Penn State and the University of Illinois Chicago has been awarded $900,000 from the U.S. National Science Foundation to employ biomedical research, clinical data, advanced artificial intelligence (AI) and mathematical modeling methods to ultimately support personalized medicine for people with Alzheimer's disease. The research team will use AI large language models to evaluate existing research on the disease and combine their results with clinical data to produce digital twin models - computer representations of patients or groups of patients - that will help the team map the trajectory of the disease as well as the impact of various treatments.
"Alzheimer's disease affects millions of Americans, but it can manifest differently and there is no single, clear cause," said Wenrui Hao, professor of mathematics in the Penn State Eberly College of Science, director of the Center for Mathematical Biology in the Huck Institutes of the Life Sciences and principal investigator on the grant. "Current practice is to identify a one-size-fits all treatment, but a more tailored approach may be particularly useful for this highly variable disease. Our digital twin approach will allow us to predict disease progression as well as assess treatment options tailored to patients, with the ultimate goal of identifying personalized treatment strategies."
The researchers will first use a large language model - a type of AI trained on a significant amount of text data to recognize patterns - to evaluate the vast quantity of already published scientific research on Alzheimer's disease, ranging from basic science to clinical outcomes. They are particularly interested in identifying networks of biomarkers, or measurable factors in a person that might indicate the cause of the disease or that influence disease risk.
Using advanced mathematical modeling tools, the researchers will combine the results of the large language model with anonymized clinical data from the Alzheimer's Disease Neuroimaging Initiative - which gathers data such as blood work, brain scans including functional MRI, and genetics and behavioral information for research use - to create digital twin models of patients with Alzheimer's disease. They will produce digital twins at the population level, combing data from groups of people to understand the overall shared physiology of the disease, as well as models of individual patients.
"A digital twin is essentially the digital copy of a patient that reflects their health state," said Rui Zhang, assistant professor of computer science and engineering at Penn State and co-principal investigator on the grant. "Because we cannot wait years to see how a real patient develops the disease in order to treat them, we can instead use a digital twin to simulate their disease progression. Eventually, we plan to explore the impacts of different treatment options for individuals and, using machine learning, to identify the most effective individualized treatments."
These strategies could include identifying factors in mid-life that could be preventable and reduce risk of later developing the disease, as well as treatment options once the disease begins to manifest, the researchers said.
"While our current work focuses on Alzheimer's disease, the underlying techniques - such as digital twin modeling and large language model-driven mathematical reasoning - have broad potential across many areas of science and engineering," said Wenpeng Yin, assistant professor of computer science and engineering at Penn State and co-principal investigator. "These approaches could transform how we integrate data, theory and simulation to accelerate discovery well beyond biomedical applications."
The team noted that this type of research requires a variety of skills and backgrounds, including mathematical modeling, fine tuning large language models and using advanced computational tools, as well as working with and predicting various types of data. The research team is also led by Lu Cheng, assistant professor in the Department of Computer Science at the University of Illinois Chicago.
"This type of workflow could be a new paradigm for understanding the body and supporting precision medicine," Hao said.
