'DeepRare AI' Developed for Rare Disease Diagnosis
A new diagnostic system called DeepRare AI has been developed to shorten the "diagnostic odyssey" for rare diseases. The system integrates clinical, genetic, and phenotypic data to deliver evidence-linked and transparent predictions, highlighting the growing role of data science in clinical diagnostics.
- The "diagnostic odyssey" for rare diseases, which affects over 300 million people worldwide, takes an average of 4.7 to 6 years from symptom onset to a confirmed diagnosis. - DeepRare was developed by a joint team from Shanghai Jiao Tong University's School of Artificial Intelligence and Xinhua Hospital, demonstrating a collaboration between computer science and clinical medicine. - The system functions using an "agentic" architecture with more than 40 specialized AI agents. These agents perform distinct tasks such as extracting symptoms from notes, analyzing genetic variants, and searching medical literature before a central host synthesizes the information. - In a test against five expert rare disease specialists, DeepRare achieved a first-attempt diagnostic accuracy of 64.4%, compared to the specialists' average of 54.6%. - Unlike AI that relies on simple pattern matching, DeepRare mimics a doctor's "slow thinking" by forming a hypothesis, verifying it against evidence, and running self-reflection loops to correct logical gaps. - When incorporating genetic sequencing data, the system's accuracy in complex cases exceeds 70.6%, outperforming the widely used international bioinformatics tool, Exomiser, which stands at 53.2%. - The technology is already in use; an online diagnostic platform launched in July 2025 has registered over 1,000 professional users from more than 600 medical and research institutions globally. - For clinicians and genetic counselors, the system provides a complete and traceable chain of evidence for its conclusions, allowing medical professionals to understand the "why" behind a diagnosis rather than just accepting a black-box prediction.
