China's DeepRare AI Hits Record in Rare Disease Diagnosis

Researchers at Shanghai Jiao Tong University and Xinhua Hospital have unveiled "DeepRare," an AI system that has set a new global benchmark for diagnosing rare diseases ^[1][3]. According to a study published in the journal Nature, the system achieved a first-attempt accuracy of 57.18% using clinical phenotypic information alone—a 24-percentage-point lead over previous models ^[2][4]. When researchers integrated genetic data, the system's diagnostic accuracy surpassed 70%, significantly outperforming established international tools like Exomiser ^[3][5]. DeepRare employs an "agentic" architecture featuring over 40 specialized AI agents that simulate the analytical processes of human clinicians ^[2][5]. Unlike traditional "black box" AI, the system follows an iterative cycle of hypothesis generation, verification, and self-reflection ^[1][3]. This iterative approach provides a traceable chain of evidence for every diagnostic conclusion, allowing doctors to verify the underlying logic rather than just viewing a final result ^[3][4]. The platform has already registered more than 1,000 professional users from 600 institutions since its online launch in July 2025 ^[3][6]. Project lead Sun Kun announced plans to establish a global AI alliance to further validate the system using 20,000 real-world cases ^[3][4]. This advancement aims to reduce diagnostic wait times for the estimated 300 million people living with rare conditions worldwide ^[5][6].