Clustering rare diseases within an ontology-enriched knowledge graph.

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Identifying shared etiological and pathophysiological aspects of rare diseases is crucial for drug repurposing.
• Current methods may not fully capture the complex relationships between rare diseases.

Purpose of the Study:

• To develop and validate a knowledge graph-based approach for clustering rare diseases.
• To facilitate drug repurposing by identifying groups of rare diseases with shared characteristics.

Main Methods:

• Extracted data on 3242 rare diseases from the NCATS Genetic and Rare Diseases Information center.
• Constructed an integrative knowledge graph enriched with biomedical data (gene ontologies, pathways, drug-target activity).
• Applied node embeddings for clustering and validated clusters using semantic similarity and gene enrichment analysis.

Main Results:

• Generated 37 disease clusters, each containing an average of 87 diseases.
• Validated clusters quantitatively using Orphanet Rare Disease Ontology semantic similarity.
• Identified highly related enriched genes within disease clusters, indicating shared biological underpinnings.

Conclusions:

• Node embeddings are effective for clustering diseases in heterogeneous knowledge graphs.
• The developed method uncovers semantically similar diseases and relevant genes.
• The findings provide a foundation for drug repurposing by enumerating connections between disease clusters and drugs.