This research introduces a simplified retriever that improves LLM accuracy in medical phenotype normalization without requiring explicit term definitions.

• Uses contextual word embeddings from BioBERT to find candidate matches
• Searches the Human Phenotype Ontology (HPO) more efficiently
• Demonstrates improved accuracy over standard LLM approaches
• Streamlines the normalization process with fewer computational resources

Why it matters: Accurate phenotype normalization is critical for medical diagnosis, research, and treatment planning. This simplified approach makes implementation more practical in clinical settings while maintaining high accuracy.

A Simplified Retriever to Improve Accuracy of Phenotype Normalizations by Large Language Models