Scientists Use ChatGPT-Like Technology to Assess Stroke Risk From Human DNA

A team from the University of Technology Sydney (UTS) identified specific genes associated with an increased predisposition to ischemic stroke, the most common type of stroke, in which a blood clot blocks blood flow to the brain.

“When someone has a stroke, especially at a young age, they often want to know whether other family members are also at risk,” said Kairui Guo, the study’s lead author. “New AI-based tools will help explain this risk to families more effectively.”

Previous research suggests that genetic factors account for approximately 38% of an individual’s stroke risk.

“Traditionally, genetics has not been considered a key factor in stroke risk assessment; however, this study shows just how strongly inherited characteristics can influence risk,” Guo said. He added that without AI, it would be impossible to process such large volumes of data at the individual level and provide personalized risk assessments.

To improve the accuracy of stroke risk predictions, the team developed a new transformer-based AI model, using the same technology that powers large language models, including ChatGPT. The researchers trained it on genetic variations to identify their associations with ischemic stroke.

The model identified 18 genetic variants associated with ischemic stroke. It highlighted key genes, including EDEM2, which is active in the cerebral cortex, and KANK2, which is expressed in whole blood, as major factors contributing to increased stroke susceptibility.

The team tested the model using two large medical datasets that together included more than 40,000 people.

The system successfully divided participants into groups with low, moderate and high genetic risk. Over time, people in the high-risk group were found to have a 21% higher likelihood of experiencing a first ischemic stroke than those classified as being at low risk.

Until now, scientists have assessed genetic risk using polygenic risk scores (PRS), a statistical tool that estimates the likelihood of developing a disease based on the combined effects of numerous genetic variations. However, this approach does not account for how different variants interact with one another or how their specific positions in the DNA sequence affect overall risk.

The new model could pave the way for personalized and targeted treatments in the future.

“AI can be used to identify links between human genes and diseases and to develop personalized risk assessments to prevent potentially life-altering or fatal events,” said Jie Lu, director of the Australian Artificial Intelligence Institute.

According to Lu, the next step is to develop a multimodal AI tool that combines genetic risk assessments with other types of data, including health and lifestyle questionnaires, scans and medical images, to assess stroke risk more accurately.

“Genetics is only one component of stroke risk. The ability to combine a genetic score with clinical information and lifestyle data, including established risk factors such as blood pressure, age, atrial fibrillation, type 2 diabetes and cholesterol levels, could significantly improve monitoring and prevention,” Lu said.