Predicting Neurodegenerative Diseases With Deep Learning
A primary goal of genetic studies on complex diseases is to build predictive models to assess an individual's disease risk based on their genome. Current approaches primarily rely on collections of individual genetic variants identified through traditional genome-wide association study (GWAS) analysis and provide limited predictive accuracy. However, genes function in the context of a complex biological system, engaging in interactions with other genes to fulfill diverse functions. Thus, accurate disease risk prediction models will need to incorporate genetic interaction effects.
Professor Chad Myers (Computer Science and Engineering) and his research group have developed a method, called BridGE, that provides a new, computational approach for discovering genetic interactions from GWAS data by incorporating knowledge of biological pathways. BridGE has been applied successfully to identify interactions for six diseases to date but has not yet been incorporated into individual disease risk prediction models. In a project called “Multi-task deep learning models for neurodegenerative disease risk prediction,” the group aims to build more accurate disease risk prediction models for neurodegenerative diseases (Parkinsons, Alzheimer’s disease, and ALS). They will leverage a state-of-the-art graph neural network deep learning approach that is specifically designed to capture interaction effects and enables joint learning of multiple human phenotypes simultaneously. Successful completion of the project will produce more accurate individual risk prediction models and new insights into the genetic basis of neurodegenerative disease.
This project recently received a DSI Seed Grant. DSI Seed Grant funds are intended to promote, catalyze, accelerate, and advance U of M-based data science research so that U of M faculty and staff are well prepared to compete for longer term external funding opportunities. Priority is given to projects that will enable or bridge applications to larger funding opportunities and/or create new cross-disciplinary or cross-system collaborations, and to those that align with at least one of the areas of the MnDRIVE initiative. Projects must align with one of the current DSI focus areas, Foundational Data Sciences or Digital Health and Personalized Health Care Delivery. This project falls under the Brain Conditions research area of the MnDRIVE initiative and the Digital Health and Personalized Health Care Delivery focus area.
Image description: (A) Disease-associated pathways in a biological system (B) predict individual’s disease risk based their pathway-level genetic interaction profile.
