Rare Disease, Real Kid: Why We're Running a Hackathon Around One Child's Genome

Why we are running an open research hackathon around the genome of one child living with Mosaic Variegated Aneuploidy, an ultra-rare condition affecting fewer than 50 people worldwide, by Sage Bionetworks

There's a particular kind of helplessness that comes with being told your child has a disease almost no one has heard of.

That's what happened to Jonathan Bracey and his family in August 2023. Their child was first diagnosed with a rhabdomyosarcoma tumor, and shortly after, with Mosaic Variegated Aneuploidy (MVA), an ultra-rare genetic condition affecting fewer than 50 people worldwide. As Jonathan puts it:

"We were always told, 'get a diagnosis and then we will know what to do to fix your son.' No one told us that if we find a condition like MVA, the plan changes, and the best that can be offered is a prayer and keeping your fingers crossed. This was not an acceptable plan!"

There was no established research pipeline, no clinical trials, no center of excellence, no patient advocacy organization: nothing built to support a family facing this diagnosis. So Jonathan and his family built the MVA Society to fill this gap.

"In creating the MVA Society, we have set the wheels in motion to find a treatment for this ultra-rare, ultra-complex condition," Jonathan says. "We are also very clear in our aims and are focusing on research as our priority. Every penny we raise goes towards research."

That urgent need for answers is what led to Rare Disease, Real Kid: The MVA Hackathon 2026, a global open research hackathon built around real genomic and clinical data from a child living with MVA. Over two months, teams anywhere in the world will work with that data on two problems: pinpointing the variant driving the child's condition, and reasoning over existing approved drugs to surface repurposing candidates worth further investigation. When we approached the MVA Society about running an open research hackathon, the response was immediate.

"The offer of running a hackathon to support this research effort was welcomed with open arms," Jonathan says. "We are extremely excited to hear what amazing ideas come out of it. It will be a 'first' for me to be involved in something as innovative and explorative as this!"

Why open science, and why now

Every variant, every phenotype, and every submission in this Hackathon connects back to one real child. There is currently no established treatment for MVA, and the MVA Society is funding focused research to bring treatments closer. For now, the options are symptom control and cancer surveillance.

Traditional biomedical research is built on large numbers. Cohorts, statistical power, addressable market: the economics of drug development assume a population large enough to justify the investment. That logic works passably for common diseases. It fails completely for a child whose condition few clinicians will ever encounter.

We don't intend the impact of this work to stop with one person or one condition. Rare diseases, as a category, are chronically underserved by traditional research pipelines because they are individually rare but collectively affect hundreds of millions of people worldwide. The methods, pipelines, and computational approaches developed here, for variant prioritization and for drug repurposing reasoning, are the kind of reusable infrastructure the rare disease research community needs more of.

That is why we are releasing the outputs openly. Participant submissions, code, and verified reports go out under CC BY 4.0, so the value of this work outlives the Hackathon and can be built on by the next team tackling the next rare disease. What is open is the research output. The child's data itself is not released under an open licence and remains under controlled access throughout.

How the data are handled

Opening a child's genome to researchers around the world, even under controlled access, is not a decision we took lightly, and it is worth being explicit about how it works.

The Hackathon protocol is approved by an ethics committee responsible for protecting the rights, safety, and welfare of people participating in research (WCG IRB, protocol #20252010). The data is shared with full and explicit parental permission, and only under controlled access. Hackathon participants must register individually and accept both the Hackathon Rules and the Data Access and Use Terms. Redistribution of the data through any channel is prohibited. Every participant must delete all their copies of the data within 30 days of the Hackathon closing, and confirm the deletion in writing.

Participants agree not to recontact the child, their family, or points of contact at the MVA Society. Any publication arising from the Hackathon must avoid disclosing private information beyond what the family has already chosen to share publicly.

A hackathon for scientists, clinicians AND the machine learning community

We built this for the machine learning and AI community as much as for clinicians and geneticists, with two tracks:

Track 1 is a computational variant-prioritization problem, well suited to ML engineers and computational biologists building models, embeddings, or LLM-driven pipelines. We score submissions automatically against the clinically confirmed answer on two metrics, rank points and F-max, with results appearing on a live leaderboard. Our scoring follows the approach used by large-scale rare disease benchmarks such as Stenton et al. (2024).

Track 2 invites AI-assisted reasoning over drug-target databases and the literature to generate candidate drug repurposing hypotheses, judged by an expert panel on scientific rigor, potential impact, innovation, and scalability. The aim is to surface candidates that merit further investigation, not to establish that any drug works.

If you work in ML, bioinformatics, or applied AI and have never worked on a rare disease problem before, we designed this to be an accessible way in: real data, clearly defined tasks, and a published scoring rubric.

What this Hackathon is trying to buy is research attention. MVA has had very little of it. The output is a prioritized set of leads for researchers to pursue, not guidance for anyone's care. Nothing produced here is a recommendation to take, prescribe, or stop a medication, and any candidate would need independent validation and formal study before it could mean anything clinically.

From millions to one

Something has shifted in the past two years that makes a hackathon like this worth running, namely the availability of AI tools capable of complex genomic analyses.

More than twenty agentic AI co-scientists have been published in the past year: Biomni from Stanford, Kosmos from Edison Scientific, Google DeepMind's AI co-scientist, Anthropic's Claude Science (released just a few weeks back), OpenScientist, among many others. The architectures differ. What they share is that analyses which used to take days now finish in an afternoon.

What these systems change is the financial investment required for serious investigation. When a literature review that once took a postdoc days takes an agent an afternoon, the arithmetic that renders a single individual "not worth researching" starts to come apart. Depth that was previously affordable only for diseases affecting millions becomes affordable for one.

Two things follow. Non-traditional researchers can contribute to research in diverse domains which is the democratizing part. And the unit of research can shrink to a single person without becoming trivial, which is the part that matters to this family.

That is the wager behind this Hackathon. We structured both tracks as problems for which these tools are well suited, and we release the results openly so the approach can be repeated.

We are not the first to try this

This Hackathon was inspired by similar efforts in the rare disease space:

The Wilhelm Foundation has been running its Undiagnosed Hackathon for years, bringing clinicians, geneticists, and data scientists together to work through undiagnosed cases in concentrated bursts. At the 2025 edition, held at Mayo Clinic, nearly 100 participants from 28 countries worked to identify the condition from ten families over 48 hours, and a bell rang six times, once for each diagnosis. The rare disease hackathon format goes back at least to 2017, when Onno Faber ran a weekend hackathon in San Francisco on his own genome after an NF2 diagnosis. About 300 people participated, the teams released their work publicly, and Faber wrote afterward about what it felt like from his side. It seeded a series of neurofibromatosis and RASopathy hackathons run by the Children's Tumor Foundation and Sage.

The Critical Assessment of Genome Interpretation (CAGI) has approached the same problem as a blinded community experiment: participants receive genomic data and make predictions about phenotype or disease relevance, which independent assessors evaluate against clinical or experimental ground truth. CAGI's Personal Genome Project challenges asked teams to connect genomes with individual phenotypic profiles, while its Rare Genomes Project challenges asked participants to identify causal variants in families with rare, genetically undiagnosed conditions. Together, these efforts turn the difficult, often subjective work of genome interpretation into something that can be measured, compared, and improved across methods. The Rare Genomes Project, based at the Broad Institute, pushes this model toward access as well as discovery. It is a direct-to-participant study that makes genomic sequencing available to families with rare and undiagnosed suspected genetic conditions regardless of where they live in the United States, while creating a research resource for diagnosis and gene discovery.

Others are approaching the same problem from different angles. Every Cure, founded by David Fajgenbaum after he repurposed an existing drug to save his own life, is building an AI platform to systematically screen approved drugs against diseases that have none. The Chan Zuckerberg Initiative's Rare As One network funds patient-led organizations, the same kind of organization the Bracey family had to build from nothing. The Undiagnosed Diseases Network International connects clinical sites across dozens of countries. The Xcelerate RARE Open Science Data Challenge, built on RARE-X, brings researchers and data scientists together around patient-contributed clinical, phenotypic, genetic, and outcome data from rare pediatric neurodevelopmental conditions. Its challenges have asked participants to identify underrecognized symptoms, predict diagnoses, and test therapeutic hypotheses turning shared data into a structured opportunity for discovery.

What happens next

The submission window opens on 25 August 2026 and closes on 24 October at 23:59 UTC. Expert judging and qualitative evaluation run through November, with winners announced on 25 November.

The Hackathon is supported by a $50,000 prize pool: $25,000 in cash from AWS Imagine Grant program, and $25,000 in Claude credits from Anthropic. The Anthropic credits are provided for use with Claude Science, intended to help winning teams deepen their computational analysis and extend their findings on MVA beyond the Hackathon itself, rather than as a cash-equivalent award.

It is organized by Sage Bionetworks in partnership with the MVA Society, Hugging Face, and BEACON (the Benchmarking, Evaluation, and Assessment Consortium for Science), a new consortium with the mission of advancing community science.

We are also grateful to the Wilhelm Foundation, whose Undiagnosed Hackathon pioneered this approach to undiagnosed rare disease, and whose example shaped how we designed this one.

On September 17th, the MVA Society is hosting the first MVA Research Conference, in London at Great Ormond Street Hospital, bringing together everyone in the world with a stake in this disease: clinicians, researchers, and families. We'll be presenting the Hackathon there too.

As Jonathan says: "Lives literally depend on finding a treatment."

Join us: https://www.synapse.org/Synapse:syn76251147/wiki/642892

The MVA Hackathon is not intended to provide general medical care, diagnosis, or professional medical advice.