An Improved Deep Semi-supervised JNMF Method for Biomarker Extraction of Alzheimer's Disease

Area of Science:

• Neuroscience
• Genetics
• Artificial Intelligence

Background:

• Imaging genetics analyzes neuroimaging and genetic data to understand brain disorders like Alzheimer's disease (AD).
• Traditional non-negative matrix factorization (NMF) methods have limitations due to linear assumptions, restricting nonlinear feature extraction from multi-omics data.

Purpose of the Study:

• To propose a novel joint-connectivity-based deep semi-supervised non-negative matrix factorization (JCB-DSNMF) model.
• To overcome limitations of traditional NMF and incorporate prior knowledge from multi-modal data for AD research.
• To identify regions of interest (ROI), risk genes, and risk SNP loci associated with AD patients.

Main Methods:

• Developed a JCB-DSNMF model integrating physiological constraints like connectivity.
• Applied the model to analyze correlations between neuroimaging and genetic data in AD patients.
• Compared JCB-DSNMF with existing NMF-based algorithms (JDSNMF, NMF).

Main Results:

• JCB-DSNMF outperformed other NMF algorithms in identifying and predicting AD-related biomarkers.
• A diagnostic model built with selected features achieved high classification accuracy (AUC=0.8621).
• Specific features like the brain region Putamen_L (AUC=0.903) and gene RALGAPB (AUC=0.924) showed high diagnostic value for early AD detection.

Conclusions:

• The JCB-DSNMF model effectively identifies biologically relevant biomarkers for AD.
• This approach enhances the potential for early AD diagnosis through accurate feature selection.
• The findings highlight the utility of integrating multi-modal data and advanced NMF techniques in neurodegenerative disease research.