A hypergraph neural network for prioritizing Alzheimer's disease risk genes

Area of Science:

• Computational biology
• Genetics
• Artificial intelligence

Background:

• Identifying Alzheimer's disease (AD) genetic architecture is crucial for understanding its pathophysiology.
• Current network methods often fail to capture complex, higher-order gene associations.
• Higher-order gene interactions are vital for complex disease mechanisms.

Purpose of the Study:

• To introduce HyperAD, a novel hypergraph neural network framework.
• To predict Alzheimer's disease risk genes by modeling higher-order gene associations.
• To overcome limitations of pairwise gene interaction modeling.

Main Methods:

• Constructing a hypergraph where functional gene sets form hyperedges.
• Utilizing a two-stage hypergraph message passing neural network.
• Extracting high-order association information from the constructed hypergraph.

Main Results:

• HyperAD significantly outperforms state-of-the-art methods in predicting AD risk genes.
• Predicted genes are enriched in AD-related biological processes.
• Validated genes show altered protein expression in AD patient brains and include novel candidate genes.

Conclusions:

• HyperAD offers a more accurate and biologically insightful approach to AD gene prioritization.
• The framework effectively models complex genetic interactions underlying Alzheimer's disease.
• HyperAD aids in unraveling the intricate genetic landscape of AD.