LLMs as Ontology Knowledge Banks
Evaluating how well language models recall structured biological knowledge
This research systematically evaluates how accurately large language models (LLMs) can memorize and reproduce structured knowledge from biological and medical ontologies.
- LLMs demonstrate variable performance when recalling concept identifiers and labels from ontologies
- Models show strong capability to memorize commonly used ontologies like Gene Ontology
- Performance varies significantly across different ontological domains and model architectures
- Results suggest LLMs can serve as knowledge repositories for certain biological classifications with proper prompting
For biologists and healthcare professionals, this research highlights both the potential and limitations of using LLMs to access structured biological knowledge without referring to external databases.