A computable biomedical knowledge system: Toward rapidly building candidate-directed acyclic graphs

Area of Science:

• Biomedical Informatics
• Health Research Methodology
• Network Science

Background:

• Systematic understanding of third-party variables is crucial for health research, often visualized using directed acyclic graphs (DAGs).
• Traditional DAG construction relies on literature review and expert knowledge, which can be inconsistent and introduce bias.
• There is a need for more systematic and automated approaches to identify and link variables of interest.

Purpose of the Study:

• To introduce an automatic approach for building networks that link variables relevant to health research.
• To develop a system that facilitates the systematic identification of confounding and mediating variables for DAG construction.
• To enhance the consistency and reduce potential biases in DAG development.

Main Methods:

• Utilized large-scale text mining of medical literature to construct a conceptual network.
• Employed the Semantic MEDLINE Database (SemMedDB), which stores concept-relation-concept triples.
• Categorized relations between concepts as Excitatory, Inhibitory, or General.

Main Results:

• Developed a computable biomedical knowledge (CBK) system (https://cbk.bjmu.edu.cn/) for accessing SemMedDB data.
• The CBK system allows direct retrieval of publications and their associated triples without SQL queries.
• Demonstrated the system's utility through three case studies, showcasing its application in identifying influencing factors and building DAGs.

Conclusions:

• The CBK system is freely available and user-friendly for identifying phenotype-influencing factors.
• It assists in rapidly building candidate DAGs for exposure-outcome variables in health research.
• This tool can significantly reduce the time and effort required for variable relationship exploration and DAG construction, leading to improved research design and interpretation.