On February 19, 2026 (Beijing Time), two landmark studies from Xinhua Hospital and Songjiang Research Institute in collaboration with other partners, were simultaneously published in Nature. The two studies achieved global breakthroughs in AI-powered diagnosis of rare diseases and gene therapy for neurodevelopmental disorders.
Professor Sun Kun’s Team Unveils DeepRare: A Globally First Traceable AI System to Tackle Rare Disease Diagnosis
A multidisciplinary team led by Professor Sun Kun and Professor Yu Yongguo from Xinhua Hospital, together with Professor Zhang Ya and Associate Professor Xie Weidi from the School of Artificial Intelligence and the Institute for Medical Artificial Intelligence at Shanghai Jiao Tong University, has developed the world’s first traceable agentic AI system for rare disease diagnosis — DeepRare.
Driven by clinical needs and enabled by collaborative research and translational implementation, the team achieved full-chain innovation from clinical demand to technological breakthrough and clinical application. DeepRare pioneers a novel “central–avatar” traceable Agentic AI architecture, transforming AI diagnosis from a “black box” into a transparent “digital consultation room,” fundamentally resolving the long-standing trust deficit caused by opaque medical AI reasoning processes.
This next-generation system surpasses traditional medical AI across three dimensions. Knowledge Integration: Breaking down medical data silos by deeply integrating and internalizing massive medical literature databases and real-world clinical case data. Diagnostic Reasoning: Moving beyond rapid “pattern-matching” to emulate physicians’ “slow thinking” through iterative cycles of hypothesis generation, validation, and self-reflection. Transparent Inference: Delivering full-process white-box reasoning, with every diagnostic conclusion accompanied by a complete evidence chain.
Benchmark testing demonstrates record-breaking performance. Using clinical phenotypic information alone, DeepRare achieved a top-1 diagnostic accuracy of 57.18%, surpassing the previous international best model by 23.79 percentage points — providing a “golden key” for rapid rare disease screening in primary hospitals lacking access to genetic testing. With the integration of genomic sequencing data, top-1 accuracy in complex cases exceeded 70.6%, significantly outperforming the widely used international tool Exomiser. Its reasoning reports earned a 95.4% approval rate from experts at Xinhua Hospital.
After six months online, the DeepRare platform has attracted more than 1,000 professional users and now serves over 600 leading medical and research institutions worldwide. The system has been internally deployed at Xinhua Hospital and is undergoing pilot testing as a “digital quality controller” within the hospital’s rare disease diagnostic workflow. Professor Sun also revealed plans to launch a Global AI Alliance for Rare Disease Diagnosis and initiate a “10,000-Case Clinical Validation Program,” aiming to complete 20,000 real-world cases to bring China’s AI diagnostic solutions to patients worldwide.
Coinciding with the publication, Nature invited expert Timo Lassmann from the University of Western Australia’s Child Health Research Centre to contribute a News & Views commentary titled “AI succeeds in diagnosing rare diseases.” He highly praised DeepRare’s multi-module collaborative architecture, noting that it mirrors the multidisciplinary team (MDT) consultation model of rare disease centers — “yet operates at a speed and scale beyond human capability.” He emphasized that the system’s traceable reasoning capacity may be its most significant contribution toward future clinical implementation, arguing that transparent systems capable of demonstrating their reasoning processes are far more likely to earn the trust of human experts than black-box models.
The study was supported by the National Key R&D Program “New Generation Artificial Intelligence” under the Sci-Tech Innovation 2030 initiative and major projects from the Shanghai Municipal Science and Technology Commission, among others.
Professor Qiu Zilong and Professor Li Fei’s Teams Report Breakthrough: Precision Gene Editing in the Brain Offers New Hope for Neurodevelopmental Disorders
Professor Qiu Zilong from the Songjiang Research Institute of Shanghai Jiao Tong University School of Medicine, together with Professor Li Fei and Associate Researcher Yang Kan from Xinhua Hospital, collaborated with Professor Cheng Tianlin’s team at Fudan University and Academician Li Jinsong’s team at the Center for Excellence in Molecular Cell Science, Chinese Academy of Sciences. Their work focuses on Snijders Blok–Campeau syndrome (SNIBCPS), a representative neurodevelopmental disorder caused by CHD3 mutations.
Targeting the high-frequency CHD3-R1025W mutation, the team first constructed a humanized knock-in mouse model that faithfully replicates patients’ core symptoms, creating a precise disease-simulation platform. They then engineered a novel adenine base editor, TeABE, capable of precisely converting mutant A·T base pairs into normal G·C base pairs — without inducing double-strand DNA breaks, thereby greatly reducing genomic instability risks compared with traditional approaches.
Following multiple rounds of cellular screening, the optimal sgRNA-editor combination with the highest editing efficiency and lowest off-target effects was identified. Delivered via tail-vein injection, the editor successfully reached multiple brain regions and achieved targeted correction with minimal bystander effects. Restoration of CHD3 protein levels in mice led to significant improvements in social avoidance, cognitive deficits, and motor incoordination, with normalized performance in three-chamber social interaction, novel object recognition, and Barnes maze tests. To ensure translational safety, the team conducted genome-wide off-target analysis using GUIDE-seq. Potential off-target editing rates were below 1% in human cells and even lower in mouse brains. Functional validation confirmed that minor mutations did not affect CHD3 protein function. Importantly, intrathecal injection of AAV9-TeABE in non-human primates (cynomolgus monkeys) demonstrated clear base-editing activity, validating cross-species feasibility and laying essential groundwork for future clinical trials.
The Xinhua Hospital team has dedicated over two decades to research in pediatric neurodevelopmental disorders, spanning genetic and environmental etiologies from basic science to clinical translation. In addition to this Nature publication on precision gene editing, Professor Li Fei’s team recently led a major project on early-life environmental risk factors and prevention systems for neurodevelopmental disorders, which won First Prize in the 2025 Chinese Medical Science and Technology Awards.
Driving High-Quality Innovation for the Future
In recent years, Shanghai Jiao Tong University School of Medicine has consistently aligned its research with national strategic priorities, strengthened organized scientific research, refined research governance models, and accelerated the translation of scientific achievements. These efforts have generated numerous influential breakthroughs, injecting strong momentum into medical science and technological advancement.
Looking ahead, the School and its affiliated hospitals will continue to respond to emerging national strategic demands, deepen innovation-driven incentive reforms, build powerful engines for original innovation, and advance into uncharted scientific frontiers. Through systematic promotion of high-quality scientific innovation and translational development, they aim to contribute sustained SJTUSM strength toward building a world-class medical school with Chinese characteristics and advancing national self-reliance in science and technology.
