AI system DeepRare diagnoses rare diseases in Nature study

An AI system aims to shorten the rare disease diagnostic odyssey

Researchers have developed an artificial intelligence system designed to diagnose rare diseases, a process that currently takes an average of five years or more for patients. The system, named DeepRare, generates ranked diagnostic hypotheses accompanied by reasoning that links back to medical evidence. It is detailed in a new paper published in Nature.

For the estimated 300 million people worldwide living with a rare disease, obtaining an accurate diagnosis is notoriously difficult. With over 7,000 recognized rare diseases, most clinicians will never encounter a given condition, leading to a long journey of specialist consultations and misdiagnoses.

How the DeepRare system works

DeepRare uses a range of specialized AI tools and knowledge sources to analyze patient information. Unlike a simple chatbot, it functions as an agentic system, meaning it can plan and execute a series of investigative steps to reach a conclusion.

The system's key output is a shortlist of potential diagnoses, each supported by traceable reasoning. This allows doctors to see the logical pathway—such as which symptoms or test results pointed to a specific condition—making the AI's suggestion verifiable and more trustworthy for clinical use.

The architecture of DeepRare involves several components working together:

• A reasoning engine that plans diagnostic investigations.
• Tools to search and analyze vast biomedical knowledge bases.
• A module that generates final reports with evidence-backed hypotheses.

The high stakes of rare disease diagnosis

The long diagnostic delay for rare diseases has severe consequences for patients and healthcare systems. During the "diagnostic odyssey," individuals often undergo unnecessary treatments and accumulate significant costs without addressing the root cause of their illness.

Most rare diseases are genetic and chronic, and many manifest in childhood. An earlier, accurate diagnosis can be life-changing, enabling access to proper management, targeted therapies, and clinical trials. It can also provide families with a crucial understanding of prognosis and inheritance patterns.

The scale of the challenge is immense. A rare disease is typically defined as one affecting fewer than 1 in 2,000 people. This scarcity of data and clinical experience is precisely what makes the problem suitable for AI, which can synthesize information from millions of case reports and genomic databases that no single doctor could ever review.

AI's growing role in complex medicine

DeepRare enters a field where AI is increasingly being tested for complex medical reasoning. Previous systems have shown promise in areas like interpreting medical images or suggesting diagnoses for more common conditions. However, rare diseases represent a frontier due to their complexity and the "long-tail" distribution of cases.

The research team emphasizes that tools like DeepRare are intended to assist, not replace, clinicians. The goal is to augment a doctor's expertise by quickly surfacing plausible options they might not have considered, effectively narrowing down a vast field of possibilities.

Other recent AI advances in medicine include systems that can operate a laboratory and design experiments, and large language models being fine-tuned for clinical note-taking and summarization. The push is toward creating AI "agents" that can actively solve multi-step problems, not just answer questions.

Challenges and the path forward

For any diagnostic AI, key hurdles remain before widespread clinical adoption. These systems must be rigorously validated in real-world settings to prove they improve patient outcomes without introducing new risks or biases. The "black box" problem—where an AI's reasoning is opaque—is also a major concern, which DeepRare attempts to address with its traceable reasoning feature.

Furthermore, integrating such tools into busy hospital workflows and electronic health record systems presents a significant technical and practical challenge. Doctors need interfaces that provide AI suggestions without disrupting their established processes.

The ultimate test will be prospective clinical trials. Researchers will need to demonstrate that using DeepRare actually shortens the time to diagnosis and leads to better care for the millions of people currently navigating the uncertainty of a rare disease.