Researchers use ChatGPT-style technology to predict stroke risk from a person’s DNA
A new AI tool that finds genetic factors linked to stroke could help doctors offer more personal ways to prevent and treat it.
Researchers have developed an artificial intelligence (AI) model that studies a person's DNA to find genetic factors that may increase their risk of stroke. Stroke is the world's second leading cause of death.
A team at the University of Technology Sydney (UTS) has found specific genes linked to a higher risk of ischaemic stroke. This is the most common type of stroke. It happens when a blood clot blocks blood flow to the brain.
“When someone has a stroke, particularly at a young age, they often want to know whether other family members are also at risk,” said Kairui Guo, the study’s corresponding author. “These new AI tools will help families better understand that risk.”
Earlier research suggests that genetics account for about 38% of a person's risk of having a stroke, according to previous research.
“In the past, genes were not a major part of how we thought about stroke risk. However, this study shows that genes can have a big impact,” Guo said. He added that without AI, it would be impossible to process so much data for each person and use it to assess their individual risk.
To improve predictions of stroke risk, the team built a new AI model using a technology called transformers. The same technology powers large language models such as ChatGPT. The researchers trained the model using genetic differences to find links to ischaemic stroke.
The model found 18 genetic variations linked to ischaemic stroke. It identified important genes, including EDEM2, which is active in the brain's outer layer, and KANK2, which is active in whole blood, as major factors linked to stroke risk.
The team tested the model using two large medical datasets that together included more than 40,000 people.
The model successfully placed people into three groups based on their genetic risk: low, medium and high. Over time, people in the high-risk group were 21% more likely to have their first ischaemic stroke than people in the low-risk group.
Until now, scientists have estimated genetic risk using polygenic risk scores (PRS). This statistical method estimates a person's chance of developing a disease by combining the effects of many genetic variations. However, it does not account for how different genetic variations affect one another or how their exact positions in a DNA sequence may change a person's overall risk.
Towards more personal medicine
The new model could help scientists develop treatments in the future that are better suited to each person and target their specific needs.
“AI can be used to find links between human genes and diseases, and build personal risk assessments that can help prevent events that could change or end a person's life,” said Jie Lu, director of the Australian Artificial Intelligence Institute.
Lu said the next step is to build a multimodal AI tool. This type of tool would combine a person's genetic risk assessment with other information, such as health and lifestyle surveys, scans and medical images, to give a better picture of their stroke risk.
“Genes are only one part of stroke risk. Combining a genetic risk score with medical and lifestyle information could improve monitoring and prevention. This information could include known risk factors such as blood pressure, age, atrial fibrillation, type 2 diabetes and cholesterol,” Lu said.