DEG-BRIN-GCN: interpretable graph convolutional framework with differentially expressed genes brain region...

Area of Science:

• Neuroscience
• Computational Biology
• Genomics

Background:

• Alzheimer's disease (AD) diagnosis and mechanism analysis are challenging due to complex molecular and brain region interactions.
• Current methods lack sufficient accuracy and biological interpretability.

Purpose of the Study:

• To develop a novel graph convolutional neural network framework (DEG-BRIN-GCN) for enhanced Alzheimer's disease diagnosis and biological interpretability.
• To identify key genes and brain regions involved in AD pathology.

Main Methods:

• Systematic analysis of transcriptomic data from 19 brain regions to identify differentially expressed genes.
• Construction of a differentially expressed gene-brain region interaction network (DEG-BRIN).
• Development of an AD classification model using graph convolutional networks (GCNs) leveraging the DEG-BRIN network.

Main Results:

• The DEG-BRIN-GCN model significantly outperformed traditional machine learning, Random-GCN, and PPI-GCN models in diagnostic accuracy.
• Identified critical brain regions (superior parietal lobule, putamen, frontal pole) and genes (VCAM1, MCTP1, HBB, CX3CR1) involved in AD.
• Implemented a novel "gene-region-pathway" interpretability analysis for cross-scale exploration of AD mechanisms.

Conclusions:

• Inter-regional molecular interaction networks are crucial for accurate AD diagnosis.
• The DEG-BRIN-GCN framework offers a powerful tool for both diagnosis and understanding AD pathological mechanisms.
• This approach provides a novel framework for exploring complex diseases like AD.