About 300 million people worldwide suffer from over 7,000 known rare diseases, yet diagnosis remains challenging because patients initially present with incomplete and heterogeneous phenotypes. We introduce HPOQuest, a training‑free framework that focuses on sequential phenotype acquisition for rare‑disease diagnosis.
Starting from a small set of observed patient phenotypes, HPOQuest keeps a probabilistic disease ranking and iteratively selects the most informative follow‑up questions to assist clinicians. Confirmed phenotypes update the disease ranking, while every response (including negatives) refreshes the candidate question pool.
Across four benchmark cohorts, HPOQuest substantially improves diagnosis from sparse initial phenotypes, achieving up to a 30‑point gain at Recall@1 and a 45‑point gain at Recall@5. These results demonstrate that sequential phenotype acquisition can markedly boost rare‑disease diagnosis when clinical evidence is limited.
Review