An AI system called DeepRare matched or beat experts at diagnosing rare diseases
In a study published in Nature, the agentic AI system DeepRare reached 79 percent diagnostic accuracy versus 66 percent for rare-disease experts, while providing traceable, evidence-linked reasoning to help shorten the years-long diagnostic odyssey.
Artificial IntelligenceGNGV Editorial Team5 min read
Researchers reported in the journal Nature on February 18, 2026, that an agentic artificial-intelligence system called DeepRare can diagnose rare diseases with accuracy that matches or exceeds human specialists. Rare diseases are individually uncommon but collectively affect hundreds of millions of people, and patients often endure what doctors call a diagnostic odyssey, averaging more than five years of referrals, misdiagnoses and unnecessary tests before getting an answer.
DeepRare pairs a large language model with a system of specialized AI agents that work in sequence: extracting a patient's symptoms, matching them against known diseases and annotating genetic variants from sequencing data. In testing, the system correctly diagnosed 69 percent of patients when given both symptoms and genetic data, compared with 56 percent for Exomiser, a widely used genetic-analysis tool. In a head-to-head comparison, DeepRare reached 79 percent accuracy versus 66 percent for rare-disease experts.
A crucial feature is transparency. Rather than producing a verdict alone, DeepRare generates evidence-based reasoning chains with verifiable references to the medical literature, which experts confirmed used correct information about 95 percent of the time. That traceability is what makes such a tool usable in real clinics, because doctors can check the system's logic instead of trusting a black box, and it could significantly reduce the time spent on literature review.
The limits are real. The results come from research datasets, not yet routine clinical deployment, and performance still depends on the quality of symptom descriptions and genetic data, which are uneven across hospitals and especially scarce in lower-income settings. The system supports clinicians; it does not replace the judgment, examination and follow-up that diagnosis requires. But for families trapped in years of uncertainty, an accurate, explainable assistant that can point doctors toward the right answer faster is a deeply hopeful prospect.
