Researchers based in China have created an artificial intelligence (AI)-based tool that combines a mixture of patient information, including genetic data, clinical symptoms, and doctors’ notes, to help diagnose rare diseases more quickly and accurately.
“Rare diseases—defined as conditions affecting fewer than 1 in 2,000 people—collectively impact more than 300 million people worldwide, with more than 7,000 distinct disorders identified to date, approximately 80% of which are genetic in origin,” the investigators wrote in the paper describing their work in Nature.
“Despite their cumulative burden, rare diseases remain notoriously difficult to diagnose due to their clinical heterogeneity, low individual prevalence, and limited clinician familiarity.”
In recent years, a significant amount of research has tried to reduce the long period, sometimes more than five years, that patients with rare diseases have to wait to get a diagnosis. Success with approaches like newborn whole genome sequencing has boosted the field, but not everyone has access to experts in bioinformatics, and the large amount of data generated by these approaches can be overwhelming for healthcare providers.
In this study, the researchers built a large language model (LLM) and surrounded it with an agentic system, a set of specialized AI agents that each do one job well, for example, extracting symptoms from text, matching symptoms to known diseases, or annotating and ranking genetic variants, among others.
The tool, which they called DeepRare, uses key symptoms to find similar cases and medical papers, analyzes genetic variants, and then creates a short list of possible rare diseases.
The team, from various research institutes and universities in Shanghai, tested the tool they created on 6,401 patient cases from hospitals and public datasets across Asia, North America, and Europe. They also compared it with existing tools, several large AI models, and the opinion of rare‑disease specialists.
Generally, DeepRare performed better than standard methods used to diagnose rare diseases and against other known AI models. Using both symptoms and genetic data, the tool was able to correctly diagnose 69% of patients in one cohort. It also beat a well-known software tool called Exomiser (score of 56%), commonly used to interpret raw genomics files and find possible disease-causing variants.
In a direct comparison with expert opinion in 163 cases, DeepRare achieved a diagnosis in around 79% of cases versus 66% for the experts.
Experts reviewing reasoning chains and references used by DeepRare agreed that it used the correct information 95% of the time.
“The clinical implications of DeepRare extend beyond diagnostic accuracy to address fundamental challenges in rare disease care delivery. The system’s ability to provide evidence-based reasoning chains with verifiable references could reduce significantly the time required for literature review and case research…. Furthermore, the system’s consistent performance across different medical specialties suggests its potential as a valuable decision support tool for non-specialist physicians who may encounter rare diseases infrequently,” concluded the authors.
“This democratization of rare disease expertise could be particularly impactful in resource-limited settings or regions with limited access to specialized care, potentially reducing healthcare disparities in rare disease diagnosis.”
