Constructing High-Fidelity Phenotype Knowledge Graphs for Infectious Diseases With a Fine-Grained Semantic...

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Knowledge Representation

Background:

• Phenotypes are crucial for disease diagnosis and clinical manifestations.
• Current phenotype knowledge graphs are coarse-grained, lacking detailed attribute information.
• There's a need for detailed phenotype representation for advanced medical AI.

Purpose of the Study:

• To propose PhenoSSU (semantic structured unit of phenotypes), a fine-grained model for disease phenotypes.
• To capture the full semantic information of phenotypes using an entity-attribute-value structure.
• To develop a hybrid strategy for automated construction of fine-grained phenotype knowledge graphs.

Main Methods:

• Developed the PhenoSSU model based on the "entity-attribute-value" structure.
• Utilized 193 clinical guidelines for infectious diseases as the study corpus.
• Introduced 12 SNOMED-CT attributes and employed a hybrid strategy with MetaMap and machine learning classifiers for knowledge graph construction.

Main Results:

• Manually constructed fine-grained phenotype knowledge graphs for 193 infectious diseases.
• PhenoSSU instances captured 89.5% of the full semantics in phenotype descriptions.
• The hybrid strategy achieved an F1-score of 0.732 for concept recognition and 0.776 for attribute value prediction.

Conclusions:

• PhenoSSU is an effective model for precise phenotype knowledge representation in clinical guidelines.
• Machine learning can enhance the efficiency of constructing PhenoSSU-based knowledge graphs.
• This work promotes a shift towards fine-grained medical knowledge engineering.