Artificial intelligence has helped Boston Children's Hospital identify new diagnoses for children with rare genetic diseases that had remained mysterious for years, even after multiple expert reviews. Researchers at the hospital's Manton Center for Orphan Disease Research partnered with OpenAI to analyze genomes from 376 children who lacked diagnoses, and the AI system successfully clarified the cause of illness in 18 cases.

How Can AI Help Solve Medical Mysteries That Doctors Missed?

The challenge in diagnosing rare genetic diseases is staggering in scope. The human genome contains roughly 20,000 protein-coding genes, and while sequencing a patient's full genome is now routine, identifying which genetic variation actually causes a person's symptoms remains extraordinarily difficult. A single genome can contain thousands of variants, and determining which one is responsible for disease requires connecting genetic data to medical literature, clinical notes, and emerging research.

This is where large language models (LLMs), which are AI systems trained on vast amounts of text data to recognize patterns and generate human-like responses, excel. OpenAI's o3 Deep Research model, one of the most advanced systems available, was able to search through patient genomes alongside clinician notes and symptom descriptions to identify potential genetic causes that human analysts might have overlooked. The key advantage: the AI system doesn't get tired and can process hundreds of cases without fatigue.

"A researcher can only spend so much time on a single case. Maybe a case remained unsolved when it came to them first, but a year later a paper was published that clarifies the link between the gene and the disease," said Suyash Shringarpure, a technical researcher at OpenAI who focuses on health applications.

Suyash Shringarpure, Technical Researcher at OpenAI

What Were the Actual Results of This Study?

The research, published in NEJM AI (the New England Journal of Medicine's AI-focused publication), analyzed genomes from 376 patients across four disease categories. The AI system identified new diagnoses for 10 patients with rare neurodevelopmental diseases, four patients with neuromuscular disorders, two children who had died suddenly without a clear cause, and two patients with early childhood psychosis illnesses.

While a 5% diagnostic yield might sound modest, researchers emphasized that this represents a significant breakthrough. These were cases that had already been analyzed multiple times by human experts and had yielded no answers. Each new diagnosis meant a family finally received an explanation for years of medical uncertainty and suffering.

One striking example is Kyra Benton, who began experiencing unusual movement patterns at age 9. After years of visiting specialists who had no answers, she underwent a tracheotomy at age 13 due to severe heart complications. She had resigned herself to never knowing her diagnosis. Then, last year, researchers from the Manton Center called to tell her that the AI analysis had identified her condition: myofibrillar myopathy, a progressive genetic neuromuscular disorder that causes muscle fibers to break down.

"Last summer, about a week before my 20th birthday, we got a call from one of the researchers at the lab. She said, 'Hi, we know it's been about 15 years, but we have some news for you,' and it kind of just blossomed from there," said Kyra Benton.

Kyra Benton, Study Participant

How to Use AI Tools for Rare Disease Diagnosis: Key Steps for Medical Teams

• Gather Complete Patient Data: Researchers provided the AI system with clinician notes about each case, detailed descriptions of the patient's symptoms, and a filtered list of genes that might be responsible for the patient's condition.
• Leverage Existing Genomes: The Manton Center routinely screens patients' genomes against newly identified genes that might provide diagnoses, but these screenings often yield no results; AI can reanalyze these same genomes with fresh perspective.
• Require Human Review: All AI outputs were reviewed by the human research team before any diagnosis was confirmed, ensuring that clinical judgment remained central to the diagnostic process.

Catherine Brownstein, the scientific director of the genetic investigations arm of the Manton Center, expressed surprise at how effectively a commercial AI system could identify diagnoses in genomes that had been analyzed many times before. She noted that the sheer volume of genetic data makes human analysis impractical for many cases.

"It's a total game changer. It got almost 5% new diagnoses, which doesn't sound like a lot, but considering how many times these had already been analyzed, that's a huge number, and each one means an answer for a family," said Catherine Brownstein, Scientific Director of the Genetic Investigations Arm of the Manton Center for Orphan Disease Research at Boston Children's Hospital.

Catherine Brownstein, Scientific Director, Manton Center for Orphan Disease Research at Boston Children's Hospital

What Do Experts Say About the Limitations and Future of This Approach?

While the results are encouraging, researchers and independent experts emphasized that AI diagnosis tools are not a panacea and require careful oversight. Seven of the 18 identified diagnoses were actually "rediscoveries," meaning a treatment team somewhere had already identified the patient's condition but had not shared that information globally. Even these rediscoveries proved valuable, as they allow researchers to connect patients with new treatments as they become available.

Adam Rodman, a doctor and expert on AI in medicine at Beth Israel Deaconess Medical Center who was not involved in the research, called the paper "an exciting demonstration of AI systems' ability to diagnose diseases when used by doctors." He noted that a 5% diagnostic yield "could serve as a significant screening tool to help speed up the reanalysis of significant backlogs of cases".

Chunhua Weng, a professor of bioinformatics at Columbia University, cautioned that LLM results still require rigorous human review. She emphasized that "the appropriate use of LLMs in diagnosis requires careful attention to trustworthiness".

She

The research team was also clear that receiving a diagnosis is often only the beginning of a patient's journey. Identifying the genetic cause of a disease does not automatically lead to treatment options, and LLMs are not intended for consumer use in diagnosing or treating diseases. Instead, these tools are designed to assist medical professionals in their work.

The Manton Center works with over 3,500 individuals globally across all 50 states who are affected by rare diseases, partnering with hospitals and health centers worldwide. This study demonstrates that commercial AI systems, when used appropriately by trained medical professionals, can help democratize access to advanced diagnostic capabilities and potentially solve cases that have stumped the medical community for years.