MedKG: enabling drug discovery through a unified biomedical knowledge graph

Area of Science:

• Biomedical Informatics
• Computational Biology
• Drug Discovery

Background:

• Existing biomedical knowledge graphs face challenges with outdated information, limited accessibility, and poor data integration.
• These limitations hinder their effectiveness in precision medicine and drug discovery.

Purpose of the Study:

• To develop MedKG, a comprehensive, continuously updated biomedical knowledge graph.
• To improve data integration, accessibility, and utility for precision medicine and drug discovery.

Main Methods:

• Integrated data from 35 authoritative sources, defining 34 node types and 79 relationships.
• Implemented a Continuous Integration/Continuous Update pipeline for data currency.
• Incorporated molecular embeddings for enhanced semantic analysis.
• Developed MedLINK, a hybrid Relational Graph Convolutional Network for disease-drug link prediction.

Main Results:

• MedKG provides a continuously updated, integrated resource with extensive node and relationship types.
• Molecular embeddings improved semantic analysis capabilities.
• MedLINK demonstrated utility in disease-drug link prediction using clinical trial data.
• A user-friendly web application with APIs and visualization tools was created.

Conclusions:

• MedKG addresses critical limitations of existing knowledge graphs, offering a dynamic and accessible resource.
• The platform facilitates precision medicine and drug discovery through enhanced data integration and predictive modeling.
• MedKG is freely available, promoting broader access for researchers.