Self-guided Knowledge-Injected Graph Neural Network for Alzheimer's Diseases.

Area of Science:

• Neuroscience
• Machine Learning
• Artificial Intelligence

Background:

• Graph neural networks (GNNs) excel at analyzing complex, irregular data structures.
• Standard GNNs require domain-specific knowledge for optimal performance in specialized fields like Alzheimer's Disease (AD) brain connectome analysis.
• Manual integration of AD expertise into GNNs is resource-intensive and requires significant specialist input.

Purpose of the Study:

• To develop a novel self-guided, knowledge-infused multimodal GNN framework for autonomous domain knowledge integration in AD research.
• To overcome the limitations of manual knowledge curation in GNN model development.
• To enhance the efficacy and interpretability of GNNs for AD analysis.

Main Methods:

• Conceptualized domain knowledge as natural language.
• Developed a specialized multimodal GNN framework to process natural language knowledge.
• Leveraged uncurated textual domain knowledge to guide the GNN's learning process.
• Compiled a literature dataset of AD publications and integrated it with real-world AD datasets.

Main Results:

• Demonstrated the framework's effectiveness in extracting curated knowledge from AD literature.
• Showcased the ability to provide graph-based explanations for domain-specific applications.
• Achieved enhanced GNN performance by utilizing extracted domain knowledge.
• Improved prediction interpretability in the context of AD.

Conclusions:

• The proposed self-guided knowledge-infused multimodal GNN autonomously integrates domain knowledge, reducing reliance on manual curation.
• This approach enhances GNN performance and interpretability for Alzheimer's Disease connectome analysis.
• The framework offers a scalable and efficient method for leveraging textual expertise in specialized machine learning applications.