LLMBDC: Zero-Shot LLM Framework for Context-Aware GO Term Clustering
LLMBDC is a novel training-free framework designed for the clustering of Gene Ontology (GO) terms. This innovative approach leverages zero-shot semantic reasoning alongside confidence scoring to group GO terms into BioDomains using only ontology data. Unlike existing tools such as REVIGO, GOSemSim, and clusterProfiler::simplify(), which rely on fixed similarity metrics, or Metascape, which employs gene-overlap measures, LLMBDC offers a more dynamic solution. Additionally, it overcomes the limitations of GO-slim's static hierarchy mappings and the subjectivity of manual curation. The comprehensive study is accessible on arXiv under the identifier 2608.00099.
Key facts
- LLMBDC is a training-free framework for clustering Gene Ontology terms.
- It uses zero-shot semantic reasoning of LLMs with confidence scoring.
- It clusters GO terms into BioDomains using only ontology information.
- Existing tools like REVIGO, GOSemSim, and clusterProfiler::simplify() use fixed similarity metrics.
- Metascape uses gene-overlap measures.
- GO-slim uses static hierarchy mappings.
- Manual curation is subjective and labor-intensive.
- The paper is available on arXiv under identifier 2608.00099.
Entities
Institutions
- arXiv