OPENAI
AI just diagnosed the undiagnosable.
The thing about rare diseases is that the diagnosis is often just as elusive as the cure.
Families can wait years – even decades – for an answer to their medical mysteries. Along the way, many end up at Boston Children’s Hospital, a renowned institution that takes on the trickiest pediatric cases. And even still, certain conditions stump the most seasoned specialists.
That is, until OpenAI o3 entered the chat.
The model excels in STEM and has been used for PhD-level scientific analyses. So Boston researchers put the tool to the test, and the results were remarkable. After countless dead-end studies, the LLM was able to give 18 families a bit of clarity.
By the numbers:
Researchers used OpenAI o3 to analyze 376 undiagnosed patient genomes.
It identified 18 new diseases across neuromuscular, neurodevelopmental, and psychiatric categories.
This means it solved nearly 5% of the unknown medical conditions it tested.
Cracking the code
DNA has around 20,000 protein-coding genes. Finding the one responsible for a child's rare illness is like finding a needle in a haystack.
Doctors can do this legwork, but it’s arduous at best. Fortunately, OpenAI’s o3 model doesn’t get tired and can cross-reference a massive data set – like the human genome – in record time.
So while AI doesn’t guarantee a cure, an accurate diagnosis is the first step towards progress.
This is giving me so much hope! - TL