Agentic Knowledge Graph Traversal in Protein-Protein Relation Grounding
Gabriel Reder, Ross King, Carl Collins, Larisa Soldatova
EPiC series in technology
Problems Identified (4)
Biomedical relation grounding: The work addresses extracting structured knowledge from biomedical texts by grounding protein-protein interaction relations to ontology terms.
LLM structured-knowledge interfacing: The abstract identifies that LLMs struggle to interface with structured knowledge representations despite their summarization strengths.
Biomedical relation grounding: The work addresses extracting structured knowledge from biomedical texts by grounding protein-protein interaction relations to ontology terms.
LLM structured-knowledge interfacing: The abstract identifies that LLMs struggle to interface with structured knowledge representations despite their summarization strengths.
Proposed Solutions (5)
LLM agents for PPI grounding: The paper investigates Large Language Model agents for extracting structured biomedical knowledge and grounding PPI relations to PSI-MI ontology terms.
Knowledge graph interaction strategies: The authors equip agents with multiple knowledge graph interaction strategies to perform PPI grounding.
PageRank-guided graph traversal: The paper evaluates PageRank-guided traversal as a graph-topology-based strategy for agentic knowledge graph interaction.
LLM agents for PPI grounding: The paper investigates Large Language Model agents for extracting structured biomedical knowledge and grounding PPI relations to PSI-MI ontology terms.
Knowledge graph interaction strategies: The authors equip agents with multiple knowledge graph interaction strategies to perform PPI grounding.
Results (3)
PageRank traversal outperforms baselines:
Extracts missed curator knowledge:
Knowledge-base structure is informative:
Research Domain
Biomedical text mining and ontology-based relation grounding