Alzheimer's Disease Knowledge Graph Enhances Knowledge Discovery and Disease Prediction.

Area of Science:

• Biomedical Informatics
• Neuroscience
• Computational Biology

Background:

• Alzheimer's disease (AD) is a progressive neurodegenerative disorder with no effective treatments.
• Knowledge graphs (KGs) offer novel approaches for drug repurposing and biomarker discovery in complex biological networks.
• This study focuses on constructing an AD-specific knowledge graph to analyze interactions between AD, genes, chemicals, and diseases.

Purpose of the Study:

• To develop an Alzheimer's Disease Knowledge Graph (ADKG) integrating diverse biomedical data.
• To identify potential therapeutic targets and diagnostic methods for AD.
• To facilitate drug repurposing and biomarker discovery for Alzheimer's disease.

Main Methods:

• Annotated 800 PubMed abstracts and used GPT-4 for data augmentation for named entity recognition (NER) and relation classification.
• Developed a data mining model integrating NER and relation classification to extract relation triplets from biomedical literature.
• Employed biomedical databases and abbreviation resolution for entity linking, constructing the ADKG with over 5,000 unique entities and millions of connections.

Main Results:

• Successfully extracted 3,199,276 entity mentions and 633,733 relation triplets.
• Constructed a comprehensive Alzheimer's Disease Knowledge Graph (ADKG) by integrating extracted data and external databases.
• Demonstrated the ADKG's utility in generating testable hypotheses and improving predictive models, outperforming others in UK Biobank data analysis.

Conclusions:

• The Alzheimer's Disease Knowledge Graph (ADKG) is a valuable resource for advancing AD research.
• ADKG aids in hypothesis generation and enhances predictive modeling for Alzheimer's disease.
• This knowledge graph holds significant potential for developing novel treatment strategies and diagnostic methods for AD.