
When you talk to those living with rare diseases, you hear the same story again and again: getting a diagnosis is most of the battle.
For the 400 million people globally living with a rare disease, the journey to a diagnosis is long, confusing, and isolating. Half of them are still waiting for answers, and it can take an average five to seven years to find them. During that time, they see specialist after specialist, undergo repeated testing, receive incorrect diagnoses, and struggle to find an explanation for their symptoms that no one seems able to piece together. Families often describe this period as a “diagnostic odyssey” – a long and exhausting search for answers.
The Catch-22 of the diagnostic odyssey is that the data needed to diagnose these diseases already exists. Patients and their caretakers are able to describe their symptoms with incredible depth and accuracy. The problem is that our current healthcare system is unable to capture and use it. Data is often disjoined, siloed, and unusable – and clinicians frequently struggle to make sense of patient symptoms.
That’s the problem we’re trying to solve.
At Probably Genetic, our goal is to build the world’s most comprehensive database for rare diseases. By combining clinical records, patient-reported information, and biological data, we can train AI systems to identify undiagnosed patients and help connect them with answers much faster – and ultimately build a strong foundation for drug development.
This is why we are honored to have been awarded up to $10 million from the Advanced Research Projects Agency for Health (ARPA-H) to end the diagnostic odyssey for good.
ARPA-H is inspired by the Defense Advanced Research Projects Agency (DARPA) and was created to fund ambitious, high-risk efforts with the potential to transform medicine. Programs built on the DARPA model have helped catalyze some of the world’s most consequential technologies, including the internet and autonomous vehicles. Through its Rare Disease AI/ML for Precision Integrated Diagnostics (RAPID) program, ARPA-H is now applying that same spirit to healthcare, investing in the idea that artificial intelligence could help end the diagnostic odyssey for patients with rare diseases.
But for AI to meaningfully change diagnosis, something essential has to come first: data.
Despite years of excitement about AI in healthcare, most efforts in this space have been constrained by the same limitation: fragmented and incomplete datasets. Electronic health records capture only part of the story. They often miss critical information like symptom onset, progression, patient experience, and morphological features – all of which can be essential clues when diagnosing rare diseases.
Without comprehensive datasets, even the most advanced algorithms struggle to detect the patterns that lead to diagnosis.
At Probably Genetic, we approach the problem differently. Rather than relying solely on clinical infrastructure, we work directly with patients and caregivers to capture the full picture of disease. That includes what physicians observe, what patients experience, and the underlying biological information that connects the two.
Our platform integrates electronic health records, patient-reported information, and genetic data into structured datasets that allow AI models to detect patterns that would otherwise remain invisible. By combining these sources of information, we’re able to capture the deep phenotypic and genotypic signals that are often missing from traditional medical records.
In other words, we’re not just applying AI to healthcare data. We’re building the data infrastructure needed for AI to actually work.
The RAPID program will significantly accelerate this effort. With ARPA-H’s support, we’ll expand our platform to reach more patients, integrate additional sources of data, and continue refining the AI models that help identify individuals who may have undiagnosed conditions.
Importantly, this work builds on what we’ve already begun. To date, Probably Genetic has collected data from more than 120,000 patients across dozens of disease areas and partnered with more than 50 patient advocacy organizations. Our AI platform allows patients to describe their symptoms in detail and submit additional evidence (photos, videos, etc.), transforming that information into structured data that can be analyzed alongside clinical records and genetic information.
This direct-to-patient approach allows us to capture information that typically never appears in traditional medical datasets, while making it easier for patients to participate in the diagnostic process.
The result is a new kind of dataset, one that reflects both the clinical and lived experience of these diseases. That dataset, in turn, can power AI models capable of detecting conditions earlier and connecting patients with the testing and care they need. And beyond just patient impact, this kind of data infrastructure will create the foundation needed for the future of rare disease drug development and precision medicine.
Over 100,000 patients have created a profile with Probably Genetic to confirm their eligibility for our genetic testing and counseling programs.