Generating Biomedical Knowledge Graphs from Knowledge Bases, Registries, and Multiomic Data.

Area of Science:

• Biomedical Informatics
• Data Science
• Knowledge Representation

Background:

• Large-scale clinical and multiomics datasets are growing, necessitating transformation into actionable information for automated reasoning.
• Barriers to data accessibility include content diversity, size, and privacy concerns, hindering effective data utilization.
• Existing systems require harmonization of goals, standards, and resources for robust data transformation.

Purpose of the Study:

• To transform multiomics datasets and electronic health records (EHRs) into compact knowledge representations within a knowledge graph (KG) structure.
• To demonstrate this data transformation within the Biomedical Data Translator ecosystem.
• To provide access to five resulting KGs and showcase their utility in answering biomedical research questions.

Main Methods:

• Leveraging knowledge graphs (KGs) for structured, machine-learning-readable knowledge representation.
• Developing methods for transforming multiomics data and EHRs into KG data structures.
• Integrating transformed data within the Biomedical Data Translator framework, ensuring privacy preservation.

Main Results:

• Successfully transformed diverse datasets (clinical trials, drug approvals, cancer, wellness, EHRs) into five distinct KGs.
• Demonstrated the preservation of individual privacy during data transformation.
• Provided access to the generated KGs via the Translator framework.

Conclusions:

• The developed methods enable the transformation of complex biomedical data into computable KGs, supporting automated reasoning.
• The KG approach facilitates biomedical research by making data more accessible and actionable while maintaining privacy.
• Further research is needed to address challenges in extracting biomedical knowledge from multiomics data.