AI System DeepRare Achieves 70% Accuracy in Rare Disease Diagnosis
DeepRare, a multi-agent AI system, demonstrated exceptional performance in diagnosing rare diseases across 2,919 conditions, achieving 70.6% accuracy with genomic data and outperforming existing clinical tools in international testing.
A new artificial intelligence system designed to diagnose rare diseases has demonstrated it can match and in some cases surpass existing clinical tools, according to research published in Nature. The system, called DeepRare, achieved a 70.6% accuracy rate in complex cases when genomic sequencing data were included, compared to 53.2% for Exomiser, a widely used international tool for genetic analysis.
Rare diseases affect more than 300 million people worldwide, with more than 7,000 distinct disorders identified to date, approximately 80% of which are genetic in origin. Despite their cumulative burden, rare diseases remain notoriously difficult to diagnose due to their clinical heterogeneity, low individual prevalence and limited clinician familiarity. Patients often experience a prolonged 'diagnostic odyssey' averaging more than 5 years, marked by repeated referrals, misdiagnoses and unnecessary interventions, all of which contribute to delayed treatment and adverse outcomes.
The study was conducted by a joint team from Shanghai Jiao Tong University's School of Artificial Intelligence and Xinhua Hospital affiliated with its School of Medicine. DeepRare was evaluated across nine datasets from literature, case reports and clinical centres across Asia, North America and Europe spanning 14 medical specialties, demonstrating exceptional performance on 2,919 diseases.
In tests using only patients' clinical phenotype information—similar to a doctor making an initial judgment based on symptoms alone—DeepRare achieved a Recall@1 of 57.18%, meaning the first diagnosis was correct more than half of cases. The performance suggests the tool could support screening in hospitals that lack routine access to genetic testing. This result outperformed the next best method by 23.79%. In multi-modal tests, DeepRare reached 69.1% compared with Exomiser's 55.9% on 168 cases.
Unlike traditional medical AI systems that primarily match symptoms to disease categories, DeepRare follows what the researchers describe as an "agentic" workflow. The system is a multi-agent system powered by large language models, integrating more than 40 specialized tools and up-to-date knowledge sources. Like a human doctor, it forms hypotheses, tests them against evidence and revises its conclusions before ranking possible diseases.
DeepRare processes heterogeneous clinical inputs, including free-text descriptions, structured human phenotype ontology terms and genetic testing results to generate ranked diagnostic hypotheses with transparent reasoning linked to verifiable medical evidence. Expert review achieved 95.4% agreement on its reasoning chains, confirming their validity and traceability.
A survey by the China Alliance for Rare Diseases covering more than 20,000 patients found that 42 percent had previously been misdiagnosed, and patients waited an average of 4.26 years before receiving a confirmed diagnosis. The rare disease knowledge landscape is rapidly evolving, with approximately 260 to 280 rare genetic diseases discovered per year, according to the International Rare Diseases Research Consortium.
The system has already been deployed on an online diagnostic platform since July 2025, with more than 600 medical institutions worldwide registered. At Xinhua Hospital, DeepRare is currently undergoing internal testing, with plans to support doctors in diagnostic cases. The research team plans to launch a global rare disease diagnostic alliance and further validate the system using 20,000 real-world cases in the coming months.