Identifying disease trajectories with predicate information from a knowledge graph.

Area of Science:

• Biomedical Informatics
• Computational Biology
• Data Science

Background:

• Biomedical literature and databases can be represented using knowledge graphs with subject-predicate-object triples.
• Disease trajectories, specific temporal sequences of disease diagnoses, are important for patient care.
• Identifying these trajectories aids in understanding disease relationships.

Purpose of the Study:

• To determine if disease sequences form trajectories using knowledge graph predicate information.
• To evaluate the added value of directional information within predicates for trajectory identification.
• To quantify performance improvements with and without directional predicate data.

Main Methods:

• Constructed knowledge graphs from biomedical literature and databases.
• Developed methods to represent indirect paths between disease-associated proteins.
• Created feature sets with and without directional predicate information.
• Compared classification performance using these feature sets.

Main Results:

• Achieved a maximum area under the ROC curve of 89.8% and 74.5% with different reference sets.
• Directional predicate information significantly improved performance by 6.5 and 2.0 percentage points, respectively.
• Demonstrated the utility of predicate information for identifying disease trajectories.

Conclusions:

• Predicates between proteins in knowledge graphs are effective for identifying disease trajectories.
• Incorporating directional information of predicates substantially enhances prediction performance.
• This method offers a novel approach to analyzing complex disease relationships in biomedical data.