MuloAD: A Multiomics Integration Model Utilizing Graph Convolutional Networks for Alzheimer's Disease Diagnosis and...

Area of Science:

• Computational biology
• Neuroscience
• Genomics

Background:

• Early and accurate Alzheimer's disease (AD) diagnosis is critical but challenging due to complex pathogenesis and limitations of single biomarker approaches.
• Integrating multiomics data offers a promising avenue for improving diagnostic accuracy and identifying novel biomarkers for AD.
• Current multiomics methods face challenges in effectively capturing cross-omics relationships for enhanced diagnostic performance.

Purpose of the Study:

• To develop and validate MuloAD, a novel graph convolutional neural network-based framework for integrating multiomics data (DNA methylation, mRNA, and microRNA expression) for enhanced AD diagnosis.
• To identify key molecular biomarkers associated with AD through feature-importance analysis within the MuloAD framework.
• To evaluate the performance of MuloAD against existing multiomics methods and single-omics approaches.

Main Methods:

• Developed MuloAD, a framework utilizing GraphSAGE for feature extraction from independent omics data.
• Implemented a view correlation discovery network to capture cross-omics relationships in a higher dimensional label space.
• Validated MuloAD using 350 samples from the ROSMAP cohort, assessing classification performance across various omics combinations.

Main Results:

• MuloAD demonstrated superior classification performance for AD diagnosis compared to existing multiomics methods.
• The framework achieved robust accuracy across different combinations of omics data.
• Feature-importance analysis successfully identified key molecular biomarkers exhibiting differential expression between AD patients and healthy controls.

Conclusions:

• Deep learning-driven multiomics integration, as exemplified by MuloAD, shows significant potential for improving the diagnosis of neurodegenerative diseases like AD.
• MuloAD provides a powerful tool for identifying novel molecular targets for AD detection and therapeutic development.
• The study underscores the complementary value of integrating multiomics data over single-omics approaches for complex disease diagnosis.