Representing a Heterogeneous Pharmaceutical Knowledge-Graph with Textual Information

Area of Science:

• Bioinformatics
• Computational Chemistry
• Data Science

Background:

• Pharmaceutical knowledge graphs integrate diverse data but often underutilize textual information.
• Conventional graph completion tasks typically exclude textual descriptions of entities like drugs and proteins.

Purpose of the Study:

• To investigate the utility of textual information for knowledge graph completion (KGC).
• To enhance KGC by generating embeddings from textual descriptions alongside graph structure.

Main Methods:

• Constructed a heterogeneous pharmaceutical knowledge graph with textual data.
• Generated joint embeddings from textual descriptions and graph structure.
• Evaluated embeddings on the link prediction task for KGC.

Main Results:

• Successfully integrated textual information into knowledge graph embeddings.
• Demonstrated the effectiveness of the approach for link prediction tasks.
• Achieved competitive results compared to existing KGC methods.

Conclusions:

• Textual information significantly enhances pharmaceutical knowledge graph completion.
• The proposed method offers a valuable tool for drug discovery and repurposing.
• Leveraging diverse data types in KGs is crucial for advancing pharmaceutical research.