On the 23rd Feb 2026, according to Shanghai Jiao Tong University, the university’s School of Artificial Intelligence, in collaboration with the affiliated Xinhua Hospital of the School of Medicine, launched the world’s first AI agent for evidence-based reasoning in rare disease diagnosis, DeepRare, providing a Chinese solution to the global challenge of rare diseases characterized by “difficult diagnosis and high misdiagnosis rates.” The related research results were published online in the international journal Nature.
DeepRare was led by Professor Zhang Ya and Associate Professor Xie Weidi of Shanghai Jiao Tong University, together with Professor Sun Kun and Professor Yu Yongguo of Xinhua Hospital. The system innovatively adopts a “central–avatar” architecture, achieving a generational leap over traditional medical AI. The platform integrates vast amounts of medical literature and clinical cases, realizing deep internalization of knowledge. Unlike conventional AI that relies on pattern matching, DeepRare applies human-like “slow thinking” through a hypothesis–verification–self-reflection process to analyze diagnostic clues. It achieves full-process evidence-based “white-box reasoning,” with each diagnostic conclusion accompanied by a complete, traceable evidence chain, effectively addressing the trust issue in AI healthcare.
Clinical data indicate that DeepRare’s first-rank accuracy using only phenotypic diagnosis reached 57.18%, which is 23.79 percentage points higher than the best international models, overcoming the challenge that “accurate diagnosis is difficult without gene testing.” In diagnostic recall, DeepRare outperformed rare disease specialists with ten years of clinical experience. When genomic sequencing data were incorporated, the first-rank accuracy for complex cases exceeded 70.61%, significantly outperforming internationally used tools. Moreover, the system-generated reasoning reports achieved 95.40% physician satisfaction, ensuring that diagnoses are “well-reasoned and fully evidence-based.”
According to the report, the DeepRare online diagnostic platform went live on July 26, 2025, attracting over 1,000 registered professional users worldwide within six months and covering more than 600 medical and research institutions. Currently, the system has been deployed and internally tested at Xinhua Hospital, where it functions as a “digital quality control officer” within the hospital’s rare disease diagnosis and treatment quality control workflow. The team is also collaborating with top domestic genetic testing institutions to achieve automated generation of clinical interpretation reports, thereby reducing the cost and barriers for genetic test interpretation.
Furthermore, the joint team has initiated a “10,000-Patient Clinical Validation Plan”, which, relying on international multi-center collaborations, will complete tens of thousands of challenging rare disease clinical validations, weaving a global intelligent diagnostic network (according to Science and Technology Daily, reporter Wang Chun).