AI/ML
Using explainable AI, genetics and neuroimaging to target the glymphatic system
About the project
The glymphatic system is a brain-wide perivascular clearance pathway which facilitates the clearance of A$\beta$, a pathological hallmark of Alzheimer’s disease, from the brain. It is regulated by the sleep-wake cycle. The function of the glymphatic system relies highly on the water channel aquaporin 4 (AQP4), which is expressed by the AQP4 gene. Identifying genetic variants of risk with respect to the function of the glymphatic system and the role of sleep on its function, and defining genetically stratified groups to identify suitable candidates for future studies will aid in the investigation of a link between sleep and neurodegeneration. However, the effects of single genetic variants are typically very small. Additionally the interaction between genes or single nucleotide polymorphisms (SNPs), as well as gene-environment interactions might impede the declaration of single genetic variants as statistically significant.
Methodological aspects
Machine learning algorithms like decision trees or tree ensembles can learn patterns in data and are therefore promising tools to address the interactions of genes, environment and lifestyle with respect to a certain trait or disease. Explainable AI (XAI) is an emerging area of research in the field of AI. Explainability gives insight into ML models, especially into the decision processes, and might therefore aid in scientific discovery. The explanations can refer to feature importance and feature effect, which indicates the direction and magnitude the feature value has on the predicted outcome, as well as the interaction of features. Different XAI methods might provide complementary information. The main methodological challenge is that XAI methods are poorly evaluated and validated thus far. We seek to evaluate and use them with respect to scientific discovery tasks.
Partners
- Division of Neurogenetics and Molecular Psychiatry, Department of Psychiatry, Medical Faculty, University of Cologne
- Helmholtz AI - Artificial Intelligence Cooperation Unit