Probably Genetic wins up to $10 million ARPA-H contract to develop an AI-powered approach to one of medicine’s most frustrating problems
For millions of people living with rare diseases, the hardest part of their illness can begin long before treatment.
It involves not knowing.
A patient may spend years visiting primary-care physicians, specialists and hospitals, undergoing genetic testing, laboratory work and imaging studies—only to leave each appointment with another unanswered question. Symptoms may be dismissed as unrelated. Tests may come back inconclusive. A suspected diagnosis may later prove wrong.
This prolonged search for an answer has become what is known as the “diagnostic odyssey.”
Now, a government-backed effort is betting that artificial intelligence (AI) could dramatically shorten that journey.
Probably Genetic, a company focused on using artificial intelligence to improve rare-disease diagnosis, has been awarded a contract worth up to $10 million from the U.S. Advanced Research Projects Agency for Health (ARPA-H). The project is part of ARPA-H’s RAPID program, which focuses on using artificial intelligence and machine learning to advance precision diagnostics for rare diseases.
The award places Probably Genetic at the intersection of three rapidly developing fields: AI, genomic medicine and patient-generated health data.
A problem measured in years
Rare diseases are individually uncommon, but collectively affect a vast population.
The Centers for Disease Control and Prevention (CDC), lists examples of rare diseases to include, “Huntington disease, spina bifida, fragile X syndrome, Guillain-Barré syndrome, Crohn disease, cystic fibrosis, and Duchenne muscular dystrophy.”
One of the challenges for physicians is that many rare disorders share symptoms with more common conditions.
Fatigue, developmental delays, seizures, gastrointestinal problems, unexplained pain or neurological symptoms can have dozens—or even hundreds—of possible causes.
For a physician, identifying the correct diagnosis can therefore resemble solving an extraordinarily complicated puzzle with missing pieces.
Genetic testing has transformed the field, but sequencing a patient’s genome does not automatically produce an answer. The difficult step is often interpreting the information: determining which genetic variants matter and whether they actually explain the patient’s symptoms.
This is where AI could play a potentially important role.
Rather than looking at a single test result in isolation, an AI system can potentially analyze large numbers of relationships among symptoms, genetics, medical history and disease progression.
Probably Genetic’s approach also emphasizes information collected directly from patients and caregivers, creating a richer picture of the patient’s experience than a conventional medical record may provide.
From fragmented to more structured
Traditional healthcare data is often fragmented.
One specialist may have information about neurological symptoms. Another may have genetic test results. A patient’s family may have observed changes over several years that never made it into the medical record.
Patients and caregivers, meanwhile, experience the disease continuously.
They see when symptoms began, how they change, what treatments have helped, what has made them worse and which seemingly minor details might actually be clinically significant.
Capturing that information in a structured form could give AI systems more material from which to identify patterns.
This offers a great advantage:
More high-quality patient data can potentially lead to better models, which can potentially lead to faster and more accurate diagnoses.
For rare diseases, where individual conditions may have very small patient populations, the ability to aggregate information across thousands of cases could be particularly valuable.
The power—and challenge—of the data flywheel
The long-term opportunity may extend beyond simply helping an individual patient receive a diagnosis.
A successful rare-disease AI platform could create a data flywheel:
More patients contribute information. That creates a larger dataset. The larger dataset can improve the ability of algorithms to recognize patterns. Better diagnostic tools could attract more patients and clinical partners, producing still more data.
Over time, the dataset itself could become a strategic asset.
Such a resource could potentially have applications beyond diagnosis, including clinical research, patient stratification and the development of new therapies.
That possibility helps explain why government agencies and investors are increasingly interested in the intersection of artificial intelligence and rare-disease medicine.
Why the ARPA-H award matters
The ARPA-H contract is significant not simply because of its potential dollar value.
ARPA-H was created to support ambitious, high-risk health research with the potential to produce transformative results. Its involvement provides Probably Genetic with both financial resources and a degree of government validation.
AI isn’t a magic diagnosis machine
There is, however, an important distinction between AI-assisted diagnosis and an AI system that can independently diagnose patients.
Medicine is not simply a pattern-recognition exercise.
Rare diseases can resemble one another. Genetic variants can be difficult to interpret. Patient-generated information can be incomplete or subjective. And an algorithm can be wrong.
A system that performs well on historical data must still demonstrate that it works reliably with real patients in real clinical environments.
That means validation will be critical.
Physicians will need to understand how the technology reaches its conclusions, how frequently it makes mistakes and how its recommendations should be incorporated into clinical decision-making.
Privacy and data governance will also remain important questions as companies build increasingly large repositories of genetic and health information.
A potentially bigger opportunity
If Probably Genetic succeeds, the impact could reach beyond shortening the wait for a diagnosis.
A reliable rare-disease data and AI platform could become part of a broader ecosystem connecting patients, physicians, researchers, genetic-testing companies and pharmaceutical developers.
For patients, the immediate promise is simple: fewer years searching for an answer.
For researchers, better-connected clinical and genetic information could make it easier to identify patient populations and understand disease biology.
For pharmaceutical companies, improved identification of patients with specific genetic conditions could eventually make clinical-trial recruitment and drug-development research more efficient.
That creates a much larger strategic question around the company.
Is Probably Genetic building merely a diagnostic tool—or could it be building an infrastructure layer for rare-disease medicine?
The road ahead
The ARPA-H contract is an important milestone, but it is not the end of the diagnostic odyssey.
The real test will be whether Probably Genetic can translate its technology and growing datasets into measurable improvements in patient diagnosis.
- Can it reduce the time required to identify rare diseases?
- Can it increase diagnostic accuracy?
- Can its models generalize across different diseases and patient populations?
- Can physicians trust and effectively use its recommendations?
- And ultimately, can the technology demonstrate enough clinical and economic value to become part of mainstream healthcare?
As these questions remain unanswered, the underlying problems remain clear. But there seems to be promise.
Conclusion
For a patient who has spent years searching for a name for their illness, even a modest reduction in that journey could be life-changing.
The potential of AI in rare diseases goes beyond just improving a computer system’s intelligence.
The goal of Probably Genetic’s ARPA-H project is to improve medicine by helping it better understand patients and the rare diseases affecting them. And in so doing, accelerate identification, care, remedies, and research.
What are your thoughts?
