RARE-PHENIX: An End-to-End AI Framework for Rare Disease Phenotyping

Phenotyping is fundamental to rare disease diagnosis, but manual curation of structured phenotypes from clinical notes is labor-intensive and difficult to scale. A new study introduces RARE-PHENIX, an end-to-end artificial intelligence framework designed to automate this process.

RARE-PHENIX integrates several stages: phenotype extraction based on large language models (LLM), standardization of phenotypes using the Human Phenotype Ontology (HPO), and supervised ranking of the most diagnostically informative HPO terms. The system was trained using data from 2,671 patients across 11 clinical sites of the Undiagnosed Diseases Network and externally validated on 16,357 real-world clinical notes from Vanderbilt University Medical Center.

The results show that RARE-PHENIX consistently outperforms PhenoBERT, a state-of-the-art deep learning model, in terms of ontology-based similarity and precision-recall-F1 metrics. Analysis demonstrated that the addition of each module (extraction, standardization, and prioritization) contributes to the improvement of the system's overall performance. RARE-PHENIX models phenotyping as a clinically aligned workflow, providing structured and ranked phenotypes that are more concordant with clinician curation.