An artificial intelligence framework for end-to-end rare disease phenotyping from clinical notes using large language models

A team of researchers led by Cathy Shyr and 10 collaborators has developed RARE-PHENIX, an end-to-end artificial intelligence framework for rare disease phenotyping from clinical notes using large language models, as announced in their February 23, 2026 arXiv submission. The research addresses the challenge of labor-intensive manual curation of structured phenotypes from clinical notes, which has proven difficult to scale in medical diagnostics. The team trained their AI system using data from 2,671 patients across 11 Undiagnosed Diseases Network clinical sites and validated it with 16,357 real-world clinical notes from Vanderbilt University Medical Center. The RARE-PHENIX framework represents a significant advancement in rare disease diagnosis by integrating three key components: large language model-based phenotype extraction, ontology-grounded standardization to Human Phenotype Ontology (HPO) terms, and supervised ranking of diagnostically informative phenotypes. Unlike existing AI approaches that typically optimize individual components of phenotyping, RARE-PHENIX operationalizes the full clinical workflow, providing structured, ranked phenotypes that are more concordant with clinician curation. The system demonstrated superior performance compared to state-of-the-art deep learning baselines across multiple evaluation metrics, achieving an ontology-based similarity score of 0.70 versus 0.58 for the baseline method. The researchers conducted ablation analyses that demonstrated performance improvements with the addition of each module in RARE-PHENIX—extraction, standardization, and prioritization—supporting the value of modeling the complete clinical phenotyping workflow rather than treating it as a single extraction task. This comprehensive approach has the potential to support human-in-the-loop rare disease diagnosis in real-world medical settings, potentially accelerating diagnosis times for patients with rare conditions who often endure lengthy diagnostic odysseys. By leveraging large language models to process unstructured clinical text and convert it into standardized, prioritized phenotypic information, RARE-PHENIX bridges a critical gap in medical AI applications.