Artificial intelligence is beginning to demonstrate measurable value in one of healthcare’s most challenging areas: diagnosing rare genetic diseases.

A new study from Boston Children’s Hospital, published in NEJM AI, shows that OpenAI’s o3 Deep Research model helped researchers identify 18 previously unsolved diagnoses among children whose conditions had remained unexplained despite years of genetic analysis.

The findings highlight how AI can support medical specialists by accelerating complex data analysis while keeping clinicians firmly in control of every final decision.

Turning Genomic Data Into Diagnoses

Advances in genome sequencing have made it relatively easy to read a patient’s DNA. Understanding which genetic variations actually cause disease remains a far more difficult challenge.

Each human genome contains around 20,000 protein-coding genes, along with millions of genetic variants. Identifying the handful of changes responsible for a rare disorder often requires reviewing years of scientific literature, comparing thousands of data points, and connecting evidence that may be scattered across multiple publications.

Researchers at Boston Children’s Hospital wanted to determine whether a commercially available AI model could assist with this process.

The team analyzed 376 unresolved patient cases, providing the model with:

clinicians’ notes,
detailed symptom descriptions,
and a filtered list of candidate genes.

The AI generated potential genetic explanations, which were then carefully reviewed by medical experts before any diagnosis was confirmed.

Eighteen New Diagnoses

The results exceeded expectations.

The research team identified 18 new diagnoses, including patients with:

rare neurodevelopmental disorders,
neuromuscular diseases,
unexplained sudden childhood deaths,
and early childhood psychosis.

While this represents roughly 5% of the unresolved cases, researchers describe the outcome as highly significant.

Many of these genomes had already been examined multiple times over several years without producing answers. Every additional diagnosis provides families with long-awaited clarity and may also improve access to future treatments or clinical trials.

AI as a Research Accelerator

Rather than replacing geneticists, the AI served as a research assistant capable of processing enormous amounts of biomedical information without fatigue.

Large language models are particularly effective at:

reviewing large collections of scientific publications,
identifying newly discovered gene-disease relationships,
connecting evidence published after a patient’s previous evaluation,
surfacing promising hypotheses for expert review.

Researchers noted that a human specialist can spend days investigating a single complex case, while an AI model can rapidly analyze multiple possibilities across hundreds of patients.

This allows specialists to focus more time on validating findings instead of searching through vast amounts of literature.

Human Expertise Remains Essential

The study also reinforces an important principle in medical AI: clinicians remain responsible for every diagnosis.

The AI did not independently diagnose patients. Instead, it generated hypotheses that were reviewed, verified, and either accepted or rejected by experienced medical professionals.

Experts involved in the study emphasized that large language models should be viewed as decision-support tools rather than autonomous diagnostic systems.

This human oversight remains critical, particularly in healthcare, where accuracy, explainability, and patient safety are paramount.

A Real-Life Impact

One of the patients featured in the study had spent nearly 15 years searching for an explanation for a progressive neuromuscular condition.

After repeated inconclusive evaluations, the AI-assisted research identified myofibrillar myopathy, a rare genetic disorder affecting muscle fibers.

Although the diagnosis did not immediately provide a cure, it finally offered the patient and family a clear explanation for years of uncertainty and may improve access to future therapies as new treatments become available.

What This Means Beyond Healthcare

The research illustrates a broader trend that extends well beyond medicine.

Many industries generate enormous volumes of structured and unstructured information that humans struggle to analyze efficiently. AI increasingly creates value by identifying patterns across large, fragmented datasets while experts retain responsibility for interpretation and decision-making.

Healthcare provides one of the clearest demonstrations of this model:

AI accelerates analysis.
Specialists validate conclusions.
Better decisions reach people faster.

As organizations continue integrating AI into knowledge-intensive workflows, the greatest impact is likely to come from systems that augment expert judgment rather than attempt to replace it.

For software teams building AI-powered enterprise solutions, this study offers another compelling example of where modern language models deliver measurable value: helping professionals navigate complexity, uncover hidden connections, and make faster, more informed decisions.

Source