DeepRare: The first AI-driven proxy diagnostic system that changes clinical decision-making in rare disease management

Rare diseases affect approximately 400 million people worldwide, accounting for more than 7,000 individual diseases, most (about 80%) of which have genetic causes. Despite the incidence, it is well known that diagnosing rare diseases is difficult. Patients have suffered through a long-term diagnostic process, on average for more than five years, often leading to sequential misdiagnosis and invasive surgery. All these delays have severe negative effects on the efficacy of treatment and the quality of life of patients. This diagnostic dilemma is largely driven by clinical heterogeneity of rare conditions, low prevalence of individual conditions, and lack of exposure by clinicians. These limitations underscore the urgent need for refined, accurate diagnostic tools that integrate a wide range of medical knowledge to detect rare situations and initiate timely interventions.

Existing diagnostic tools and their limitations

Diagnosing rare diseases widely relies on professional bioinformatics tools such as benzoprotein, a platform for processing human phenotypic ontology (HPO) terms and PubCaseFinder, which can identify and align with similar clinical cases in the medical literature. These methods mainly utilize structured clinical terms and historical case records. Meanwhile, recent advances in large language models (LLMS), including general-purpose GPT models and medically trained versions such as Baichuan-14b and Med-Palm, have begun to contribute to the diagnostic process by effectively managing multimodal clinical data. Despite these developments, existing approaches often face limitations. Traditional bioinformatics tools often lack the adaptability to keep pace with emerging medical knowledge. At the same time, universal language models may not adequately capture the nuances inherent in rare disease phenotypes and genotypes, resulting in suboptimal performance.

Introduction to DeepRare diagnostic system

Researchers from Joao University in Shanghai, researchers from Xinhua News Agency, affiliated with the Shanghai Artificial Intelligence Laboratory of Shanghai Heqiao University School of Medicine, Harvard Medical School introduced the first rare rare disease LLM-driven diagnostic platform, Deeprare. The system represents the first proxy diagnostic solution specifically designed to identify rare diseases, effectively integrating high-level language models with a comprehensive medical database and dedicated analytical components. DeepRare’s architecture is built on a three-layer hierarchical design inspired by the Model Context Protocol (MCP). The core of the central host server enhanced with long-term memory banks is powered by state-of-the-art LLM, which orchestrates the entire diagnostic workflow. The center host is a number of specialized analytical proxy servers, each designated to perform targeted diagnostic tasks such as phenotype extraction, prioritization of variants, case search and comprehensive clinical evidence synthesis. The outermost layer includes strong network-scale external resources, including the latest clinical guidelines, authoritative genomic databases, extensive patient case repositories, and peer-reviewed research literature, providing critical reference support.

Workflow of DeepRare diagnostic system

The DeepRare diagnostic process begins when the clinician enters patient data, free text clinical descriptions, structured HPO items, genome sequencing data in variant call format (VCF) or combinations. The central host systematically coordinates these proxy servers to retrieve relevant clinical evidence from external sources that are precisely tailored to each patient’s medical profile. Subsequently, preliminary diagnostic hypotheses will be generated and iteratively refined through self-reflection mechanisms, in which the host continuously evaluates and verifies emerging hypotheses through supplementary evidence collection. This iterative process effectively minimizes potential diagnostic errors, greatly reduces false diagnosis, and ensures that conclusions remain well based on verifiable medical evidence. Ultimately, DeepRare produces a list of ranked diagnostic candidates, each with explicit support from transparent and traceable reasoning chains that refer directly to authoritative clinical sources.

Evaluation results and benchmarks

In a rigorous cross-center assessment, DeepRare demonstrates superior diagnostic accuracy in clinical institutions, public case registries, and eight benchmark datasets in Asia, North America and Europe. The combined data set contains 3,604 clinical cases, representing 2,306 rare diseases in 18 medical specialties, including neurology, cardiology, immunology, endocrinology, genetics, and metabolism. DeepRare showed great diagnostic advantages, with the highest overall accuracy of diagnostic recalls at 70.6% when integrating phenotype (HPO term) and genetic sequencing data. This result exceeds the baseline diagnostic model and the simultaneous evaluation of alternative agents and LLM methods. Specifically, Exomiser’s recall rate was 53.2% compared to the second best approach, and Deeprare showed a significant improvement of 17.4 percentage points. Furthermore, in a multimodal clinical protocol containing genomic data, DeepRare’s accuracy significantly increased from 46.8% (using phenotypic data only), highlighting its proficiency in the comprehensive comprehensive patient information for accurate diagnosis.

Clinical validation and availability

Dark clinicians involved in 50 complex cases affirmed their diagnostic reasoning for dark clinician evaluation, achieving expert consistency in clinical effectiveness and edibleness of 95.2%. Physicians recognize their efficiency in producing accurate and clinically relevant references, which greatly reduces diagnostic uncertainty. For practical clinical integration, DeepRare can be accessed through a user-friendly web application that enables structured input of patient data, genetic sequencing files, and imaging reports.

Key Highlights of Deeprare

• DeepRare introduces the first comprehensive proxy AI diagnostic system, which is clearly tailored to rare diseases, integrating state-of-the-art language models, dedicated analytical modules and an extensive clinical database.
• It adopts a hierarchical architecture that includes a central host server and multiple analytical proxy servers to ensure systemic and traceability diagnostic processes.
• Compared with traditional bioinformatics tools and existing large language model systems, DeepRare’s diagnostic accuracy (70.6% recall diagnosis) was compared with traditional bioinformatics tools and existing large language model systems.
• The integration of phenotype and genomic data significantly enhances diagnostic recalls, highlighting the system’s powerful multimodal analytical capabilities.
• Expert evaluation shows that the protocol rate for the effectiveness and clinical relevance of the DeepRare transparent reasoning process is 95.2%, emphasizing its reliability in a real-world clinical environment.
• User-friendly web applications facilitate practical clinical integration, allowing comprehensive patient data entry, symptom refinement and automated clinical report generation, benefiting healthcare professionals directly.

Conclusion: Convert rare disease to dark color

In summary, this study represents a transformative advance in the diagnosis of rare diseases, significantly addressing historical diagnostic challenges through the introduction of DeepRare. By combining complex language modeling techniques with specialized clinical analytical agents and extensive external databases, DeepRare greatly improves diagnostic accuracy, reduces clinical uncertainty and accelerates timely intervention in care for patients with rare diseases.

Check Paper. All credits for this study are to the researchers on the project. Also, please feel free to follow us twitter And don’t forget to join us 100K+ ml reddit And subscribe Our newsletter.

Sana Hassan, a consulting intern at Marktechpost and a dual-degree student at IIT Madras, is passionate about applying technology and AI to address real-world challenges. He is very interested in solving practical problems, and he brings a new perspective to the intersection of AI and real-life solutions.