The average rare disease diagnosis takes six years, with some waiting decades. More than 10,000 rare diseases affect around 30 million Americans, costing the US health system an estimated $1 trillion a year, despite only 5% having an approved treatment.
Sage Bionetworks and the Advanced Research Projects Agency for Health (ARPA-H) are partnering to develop the Rare Disease Data Commons (RDDC), a central data repository and benchmarking platform for ARPA-H’s new Rare Disease AI/ML for Precision Integrated Diagnostics (RAPID) program.
RAPID is awarding Sage Bionetworks up to $28 million over 4.5 years to advance the program’s goal to transform rare disease diagnosis, enabling patients to receive answers sooner while reducing costs for families and the health system.
ARPA-H’s RAPID program is split into four technical areas:
- Assembling a large-scale longitudinal clinical dataset from electronic health records.
- Collecting patient-reported data and other modalities such as genomics, imaging, video, voice, and wearable data.
- Building the platform that ingests data from the first two areas, harmonises it, governs access, and benchmarks the diagnostic AI built on top of it.
- Validating tools in clinical settings.
Patient experience partners will work across all four areas throughout the program.
The primary obstacle the RDDC addresses is the shortage of usable data. Rare disease data sits in separate institutions, coded differently, and governed under terms that must be renegotiated every time a new research group requests access.
Robert Allaway, director of Rare Disease at Sage Bionetworks and Principal Investigator (PI) for the project, said, “Rare disease data are locked into islands. A foundation typically funds one disease, collects one data type, and the result doesn’t connect to what the next foundation collected for the next disease. We proposed a platform that connects these islands into an archipelago, solving data challenges shared across the rare disease community once instead of ten thousand times.”
The RDDC will have five connected modules that bring in data from other RAPID teams and outside contributors, make it comparable across sources by building on tools created for the ARPA-H Biomedical Data Fabric Toolbox, and set rules for who can use it and how. Where data cannot be centralised, RDDC supports federated integration and interoperates with other trusted research environments instead of becoming another silo.
Milen Nikolov, director of Product Innovation at Sage Bionetworks and co-PI for the project, said, “People living with rare diseases often drive their care odyssey, working closely with clinicians and researchers to gain insight into their diagnosis, connect to a community of others with similar conditions, and understand their prognosis and treatment options. The RDDC modules advance breakthrough technology – translating disjoint, multi-modal datasets into reliable information usable by patients, researchers and clinicians – throughout this odyssey and across all rare diseases.”
Jineta Banerjee, associate director of Advanced Data Analytics at Sage Bionetworks and co-PI for the project, said, “For rare diseases, where no individual dataset may ever be large enough, our greatest opportunity is to learn across boundaries of individual rare diseases. In the Rare Disease Data Commons, we aim to meaningfully connect datasets from multiple rare diseases to allow advanced computational methods to uncover shared and latent patterns that would help improve diagnosis or identify novel therapeutic opportunities.”
The most publicly visible deliverable is the Rare Disease Benchmark Dataset: a versioned, accessible resource designed to provide a common, credible basis for evaluation to researchers and developers working on diagnostic AI.
Today, groups developing rare disease AI methods often evaluate on proprietary data, making it difficult to compare results or know which methods are worth clinically pursuing. The benchmark would enable standardised evaluation across the field while protecting patient privacy through de-identification, governed access, and appropriate data-use controls.
Allaway said, “It’s incredibly challenging to make apples-to-apples comparisons between AI tools for rare diseases. A group builds a model, reports on its own cohort, and there is no way to tell whether the next method is better or just due to differences in the test data. We are building a place where the evaluation is the same for everyone.”
Luca Foschini, president and CEO of Sage Bionetworks, said, “About one in ten Americans lives with a rare disease; many of them are still searching for answers, and they have been waiting a long time for infrastructure built to their scale. Building in the open while protecting privacy is a complex undertaking, but it’s the only way we can create trust and share learnings for the benefit of all patients, rare disease and beyond.”
Sage Bionetwork develops and operates Synapse, one of the NIH-listed generalist repositories, managing more than 4.5 petabytes of biomedical research data. The RDDC builds on 15 years of Sage’s experience developing governed data commons for sensitive biomedical data. Sage is building the Rare Disease Data Commons with Cirro Bio, Netrias, the Gyori lab at Northeastern University, the Alsentzer lab at Stanford University and the Undiagnosed Diseases Network Foundation.
