DuAL-Net: A Dual-Network Approach for Alzheimer's Disease Risk Prediction Using APOE-Centered Regional Whole-Genome...

Area of Science:

• Genomics
• Computational Biology
• Neuroscience

Background:

• Alzheimer's disease (AD) prediction from genomic data is complex due to high dimensionality and intricate genetic variant relationships.
• Existing methods struggle to effectively integrate diverse genomic information for accurate risk assessment.

Purpose of the Study:

• To introduce DuAL-Net (Dual Approach Local-global Network), a hybrid framework for prioritizing single-nucleotide polymorphisms (SNPs) associated with Alzheimer's disease.
• To evaluate the performance and generalizability of DuAL-Net in AD risk prediction using large-scale genomic datasets.

Main Methods:

• Developed DuAL-Net, a hybrid framework combining local genomic window analysis and global annotation-based modeling.
• Applied DuAL-Net to 14,094 SNPs in the APOE region from ADNI and ADSP cohorts (n=1,050) using nested 5-fold cross-validation.
• Validated the framework on an independent ADSP cohort (n=5,570) to assess generalizability.

Main Results:

• DuAL-Net achieved an AUC of 0.698 for the top 100 ranked SNPs in the primary cohort, significantly outperforming bottom-ranked SNPs (AUC=0.479).
• Independent validation demonstrated strong generalizability, with top SNPs yielding AUC=0.686 versus 0.516 for bottom SNPs.
• The framework successfully identified known AD risk variants like rs429358 and rs7412.

Conclusions:

• DuAL-Net offers a robust and generalizable approach for Alzheimer's disease risk prediction by integrating local and global genomic information.
• The framework's ability to prioritize biologically relevant SNPs enhances the potential for targeted genetic risk assessment in AD.
• DuAL-Net represents a significant advancement in leveraging complex genomic data for predicting Alzheimer's disease susceptibility.