Heterogeneous Graph Representation Learning Framework for Resting-State Functional Connectivity Analysis

Area of Science:

• Neuroscience
• Computational Biology
• Medical Informatics

Background:

• Brain functional connectivity analysis is crucial for understanding brain development and disorders.
• Existing methods struggle to capture complex correlations and heterogeneous information within brain subnetworks.

Purpose of the Study:

• To develop a novel framework, BrainHGL, for constructing and analyzing high-order heterogeneous functional brain networks using meta-paths.
• To improve the understanding of relationships between subnetwork interactions and mental diseases.

Main Methods:

• Proposed a Heterogeneous Graph representation Learning framework (BrainHGL).
• Employed meta-path encoding for topological information, meta-path interaction for association patterns, and meta-path aggregation for fusion.
• Utilized Nanjing Medical University and Autism Brain Imaging Data Exchange (ABIDE) datasets.

Main Results:

• Demonstrated effectiveness in classifying major depression disorder (MDD), bipolar disorder (BD), and autism spectrum disorder (ASD).
• Provided insights into critical brain subnetworks, regions, and functional pathways.
• Identified disease subtypes consistent with prior neuroscientific findings.

Conclusions:

• BrainHGL offers a powerful approach for analyzing heterogeneous functional brain networks.
• The framework enhances disease classification and interpretability in neuroscience.
• This work pioneers the formulation of heterogeneous brain networks for mental disease research.