Deep-learning AI could predict Alzheimer’s risk
Researchers have developed an artificial intelligence model that can predict someone’s risk of developing Alzheimer’s Disease more accurately than previous models.
The model uses genetic information to predict an individual’s risk of predicting Alzheimer’s Disease with 70% accuracy for East Asian populations.
Alzheimer’s Disease is a fatal disease that affects more than 50 million people worldwide. Symptoms usually appear later in life, and can include cognitive dysfunction, the loss of brain cells and progressive memory loss as well as impaired movement, reasoning, and judgment.
Since Alzheimer’s Disease is heritable, it should be possible to predict which people are at greater risk of developing it in their lifetime. However, there are multiple genetic risk factors linked to the disease, so accounting for the complex effects of these is hard to capture.
Now, a research team led by Hong Kong University of Science and Technology President and Morningside Professor of Life Science, Professor Nancy Ip, and the Chair Professor and Director of HKUST’s Big Data Institute, Professor Chen Lei, have used AI to examine multiple genetic risk factors for AI.
Analysing genetic risk factors
AD is usually diagnosed after symptoms develop, using a combination of tests including cognitive assessments and brain imaging. While interventions are becoming available to slow the progression of the disease, these are more effective the earlier they are used, so being able to screen for high-risk individuals could have a significant impact.
One genetic variant associated with AD, called APOE-ε4, has been approved by the US Food and Drug Administration for screening. However, this variant alone cannot provide accurate Alzheimer’s Disease prediction, especially across different populations. For example, for the Chinese population, APOE-ε4 screening only has an accuracy of around 60%. This means many high-risk individuals without this genetic variant will be missed.
In a paper published in Communications Medicine, the team used an AI technique called deep learning to analyse patient data for a mix of genetic risks factors, including APOE-ε4, for developing Alzheimer’s Disease. The team trained the model on DNA data collected in collaboration with doctors and clinics, as well as data from existing studies conducted in non-Asian populations.
The team combined traditional statistical analyses of multiple disease-associated genetic risk factors in the data with deep learning to capture ‘non-linear’ effects of individual factors. These are impacts that may not be logical based on what is currently known. An example of non-linearity is ‘tipping points’ in the climate, where adding more heat to the atmosphere causes suddenly much stronger impacts than the same amount of extra heat had previously.
For Alzheimer’s Disease, this means genetics that affect different biological processes, such as immune responses and neuronal functions, impact individual people differently. Deep learning models therefore provide additional information about the contribution of specific biological processes to the risk of individuals developing Alzheimer’s.
Being able to find these patterns in the genetic data from Alzheimer’s Disease patients meant the new model improved the accuracy of prediction for East Asian population to over 70%. As well as identifying high-risk individuals, the model also allowed the researchers to group the disease risk into different subtypes, such as those with and without the APOE-ε4 variant.
Towards clinical confidence
While the improvements in the predictive power for Alzheimer’s Disease are remarkable, there is still some work to do before it can be used clinically with confidence. To achieve the more than 85% accuracy needed, Ip says there is room for further optimisation of the deep learning model, particularly by incorporating more genetic and Alzheimer’s biomarker information, and recruiting more patients from multiple clinical centres to train and validate the model.
This process could be streamlined by making gathering patient data efficiently and securely easier, while conforming to ethical guidelines, privacy regulations, and informed consent procedures.
Even when the test reaches its full potential though, the team caution that the responsibility for accurate diagnosis needs to be shared among multiple stakeholders, including the developers of the AI system, healthcare providers, and regulatory bodies.
But as an added tool, being able to identify potentially high-risk individuals would allow clinicians to monitor them with a range of tests and potentially provide personalised therapy based on their genetic profile.
AI in healthcare
This is not just an academic exercise. Ip said: “By identifying individuals with a high risk of Alzheimer’s Disease, we can utilise newly developed blood protein biomarker tests or brain imaging examinations to detect the disease before symptoms become apparent.” She also notes that while drugs are becoming available, there are many lifestyle factors that can affect the development of Alzheimer’s Disease.
She added: “Individuals with a high risk of developing Alzheimer’s Disease are recommended to promote a healthier lifestyle by making positive changes in diet, exercise, and sleep habits, while limiting or avoiding smoking and excessive alcohol consumption, which can potentially reduce the risk of developing Alzheimer’s Disease and other forms of dementia.”
Ultimately, the methods used by the team are transferrable to a diverse range of common diseases that are influenced by multiple genetic risk factors, such as cardiovascular diseases, certain types of cancer, neurodegenerative disorders, autoimmune diseases, and metabolic disorders. Ip said: “It is important to note that the list is not exhaustive and is potentially growing, as the potential applications of this method continue to expand with ongoing research and advancements in the field.”
Lei added: “This study exemplifies how the application of AI to the biological sciences can significantly benefit biomedical and disease-related studies. Our research also highlights how AI can elegantly, efficiently, and effectively address interdisciplinary challenges. I firmly believe that AI will play a vital role in various healthcare fields in the near future.”