The key facts

  • The transformer-based AI model identified 18 genetic variants associated with ischemic stroke; key genes are EDEM2 and KANK2.
  • Across two datasets (more than 40,000 people), the risk of a first ischemic stroke in the high-risk group was 21% higher than in the low-risk group.
  • Genetics explains about 38% of individual stroke risk; the next step is a multimodal tool combining genetics with clinical data and lifestyle.

New AI Model for Assessing Genetic Stroke Risk

Scientists at the University of Technology Sydney (UTS) have developed an artificial intelligence model that analyzes a person's DNA to identify genetic risk factors for stroke — the second most common cause of death worldwide. The team identified genes associated with an increased predisposition to ischemic stroke — the most common form of the disease, in which a blood clot blocks blood flow to the brain.

Technology and Results

The model is built on transformers — the same technology that underlies large language models, including ChatGPT. It was trained on genetic variations to establish their association with ischemic stroke and identified 18 associated variants, pointing to the key genes EDEM2 (active in the cerebral cortex) and KANK2 (in whole blood) as the main predisposition factors. The model was tested on two large medical datasets covering a combined total of more than 40,000 people. The system divided participants into low, medium, and high genetic risk groups; over time, people in the high-risk group turned out to have a 21% higher probability of a first ischemic stroke than the low-risk group.

Difference from Previous Approaches

Previously, genetic risk was assessed using polygenic risk scores (PRS) — a statistical tool that calculates the probability of disease based on the combined effect of many variations. However, this approach does not account for how variants interact with each other and how their position in the DNA sequence affects overall risk. According to previous research, genetics explains approximately 38% of individual stroke risk. Lead author of the study Kairui Guo noted that genetics has traditionally not been considered a key factor in stroke risk assessment, but the study shows a significant influence of hereditary traits; processing such a volume of data at the individual level would be impossible without AI. “When a person has a stroke, especially at a young age, they often want to know whether other family members are at risk,” Guo said, adding that new AI tools will help families better understand this risk.

The Path to Personalized Medicine

Director of the Australian Artificial Intelligence Institute Jie Lu stated that AI can be used to find links between genes and diseases and to build personalized risk assessments that help prevent a potentially life-changing or fatal event. According to him, the next step is a multimodal AI tool that combines genetic assessment with other data: health and lifestyle surveys, scans, and images. “Genetics is only one component of stroke risk. The ability to combine the genetic score with clinical information and lifestyle data, including recognized risk factors such as blood pressure, age, atrial fibrillation, type 2 diabetes, and cholesterol levels, could significantly improve monitoring and prevention,” Lu noted.