New DeepRare AI Outperforms Human Experts in Diagnosing Rare Genetic Disorders

Researchers from Shanghai Jiao Tong University and Xinhua Hospital have unveiled DeepRare, a first of its kind artificial intelligence designed to solve the world's most complex medical mysteries. Published in the journal Nature in February 2026, this never before seen system mimics the slow thinking processes used by human clinical experts.

Agentic Architecture and Reasoning

The system operates as an agentic network, coordinating over 40 specialized AI agents to analyze genetic data, clinical symptoms, and doctor notes. Unlike traditional models that offer a single prediction, DeepRare utilizes a cycle of hypothesis and verification to ensure every diagnosis is backed by traceable evidence. This mimics the iterative reasoning of a medical consultant. In rigorous benchmarking trials, the tool demonstrated a 79 percent diagnostic success rate when faced with challenging real-world cases. This significantly outperformed human specialists, who averaged a 66 percent accuracy rate in the same study. The AI also surpassed standard bioinformatics tools like Exomiser by a wide margin, correctly identifying diseases where other software failed.

Global Training and Data Integration

The researchers trained the model using a massive dataset of 6,401 patient cases spanning Asia, North America, and Europe. By integrating the Human Phenotype Ontology with genomic sequencing files, the system can pinpoint disease-causing variants that traditional methods often miss. This global dataset ensures the tool remains effective across different populations and medical specialties.

One of the most significant breakthroughs of DeepRare is its transparency. The system provides a reasoning chain that clinicians can review, citing specific medical literature to justify its conclusions. Experts who reviewed these chains agreed with the logic and references more than 95 percent of the time, reducing the fear of AI hallucinations.

By automating the labor-intensive process of literature review and variant interpretation, DeepRare aims to end the years-long wait many patients endure for a diagnosis. This democratization of expertise could prove vital for regions with limited access to rare disease specialists, potentially reducing global healthcare disparities.