Integrating and formatting biomedical data as pre-calculated knowledge graph embeddings in the Bioteque.

Area of Science:

• Bioinformatics
• Computational Biology
• Data Science

Background:

• Biomedical data is growing rapidly, posing integration challenges for comprehensive analysis.
• Existing data lacks a unified framework, hindering simultaneous consideration of multiple biological event perspectives.

Purpose of the Study:

• To introduce Bioteque, a large-scale resource of pre-calculated biomedical descriptors.
• To demonstrate Bioteque's utility in analyzing high-throughput data and predicting drug responses.
• To showcase the seamless integration of Bioteque descriptors into machine learning workflows.

Main Methods:

• Constructed a knowledge graph with over 450,000 biological entities and 30 million relationships.
• Integrated and harmonized data from over 150 diverse sources.
• Generated pre-calculated descriptors for 12 biological entity types and 67 association types.

Main Results:

• Bioteque descriptors facilitate assessment of protein-protein interactome data.
• Descriptors aid in predicting drug response and identifying repurposing opportunities.
• Off-the-shelf use of Bioteque descriptors in machine learning tasks shows no performance loss.

Conclusions:

• Bioteque provides a processed, tractable, and optimized assembly of public biomedical knowledge.
• The resource simplifies complex biomedical data analysis and supports downstream applications.
• Bioteque enhances the usability of biomedical data for research and drug discovery.