What is this challenge, exactly?
Every person's DNA contains millions of tiny variations, and almost all of them are harmless — just part of what makes each of us unique. But every so often, one of those variations disrupts something important, and the result is a disease. For many rare diseases, doctors and researchers already know a lot about the symptoms a condition causes, but they don't yet know exactly which genetic variation is responsible. Finding that answer is often the first step toward a real diagnosis, and eventually, a treatment.
This is where you come in. We're releasing real, de-identified genetic and clinical data — shared with full, explicit permission from the family it belongs to — and asking the global research community to help make sense of it. You don't need to be a geneticist to contribute meaningfully: this challenge welcomes data scientists, machine learning engineers, clinicians, students, and citizen scientists alike. If you can work with data and think carefully about a problem, you can help here.
The work is split into two tasks, described in plain terms below.
Two Tracks
Track 1 — Variant Prediction
In plain terms: you'll be given a list of genetic differences found in one patient, along with a description of their symptoms, and asked to guess which specific difference is the one actually causing their condition. Think of it like being handed a long list of suspects and a description of a crime, and asked to identify who's actually responsible, using clues from biology instead of an alibi.
Your guesses are automatically checked against the answer researchers have already confirmed through clinical testing, and your score updates on a live public leaderboard — so you can see immediately how well your approach worked and try to improve it.
What you submit: a CSV file listing your predicted variant(s), with columns p_id, c_id, chr, bp, disease — full definitions and examples are on the "How to Submit" tab.
Track 2 — Drug Repositioning
In plain terms: once you know (or have a strong guess at) what's biologically going wrong, the next question is: does a medication that already exists — approved for some other condition entirely — happen to work on this problem too? This is a real, established strategy in medicine: it's often far faster and cheaper to repurpose a drug that's already proven safe than to invent a brand-new one from scratch.
For this track, you'll write up your reasoning and proposed candidate medication(s) in a short report. There's no automatic scoring here — instead, a panel of independent expert judges (researchers, clinicians, and patient advocates) will read every submission and evaluate it on its scientific merit and real-world potential to help.
What you submit: a written report (PDF or Markdown) — no spreadsheet or code required, though you're welcome to include either as supporting evidence.
Prizes
$25,000 total prize pool — AWS-sponsored.
Judging Criteria
Submissions are evaluated across five weighted dimensions over a ~2–3 month judging window. Here's what each one actually means:
Did you identify the correct variant (Track 1) or propose a scientifically sound repositioning candidate (Track 2)?
If this were used in the real world, how much difference could it make to a patient's diagnosis, treatment, or quality of life?
Did you bring a genuinely creative angle, method, or tool to the problem, rather than a routine or purely off-the-shelf approach?
Could your approach realistically be applied to other patients, other diseases, or larger datasets — not just this one case?
How well did your team work together, document your process, and communicate your reasoning to others?
How to Participate
No prior hackathon experience needed — here's the whole process from start to finish.
-
1
Request dataset access
Because this data comes from a real patient, access isn't automatic. You'll fill out a short form on the dataset page confirming you understand and agree to the usage rules (things like not sharing the data further and deleting it once the challenge ends). A real person reviews every request, so approval isn't instant — plan to request access a few days before you want to start working.
-
2
Choose your track
Decide whether you want to try identifying the causal variant (Track 1), propose a repurposed medication (Track 2), or take on both — many teams do. You don't need to finish one to start the other.
-
3
Build & submit
Work through the data at your own pace, using whatever tools, software, or methods you prefer. When you're ready, submit your results: for Track 1, that's a simple data file; for Track 2, that's a written report. Full instructions with examples are on the "How to Submit" tab.
Ready to compete?
Questions and discussion are welcome in the HF Discussions page.
Track 1 Leaderboard
Rankings update automatically each time a valid submission is evaluated. Only the best-scoring submission per team is shown.
The leaderboard will go live once submissions open.
Check back after the challenge launch.
How to Submit
Track 1 — Variant Prediction
-
1
Prepare your file
Create a CSV file with one predicted variant per row. Required columns:
⬇ Download submission template (.csv)Field Type Description p_idstring Protein-level change (e.g. p.A571,p.A5H)c_idstring Coding sequence-level change (e.g. c.765A>T)chrstring Chromosome (e.g. chr1)bpinteger Base-pair position (GRCh38) diseasestring Free text — disease/condition associated with the predicted variant -
2
Sign in & upload
Sign in with your Hugging Face account on the leaderboard Space, then upload your file using the submission form. Only one active submission per team at a time.
-
3
See your score
Your submission is evaluated automatically. Precision and recall against the held-out validation set appear on the leaderboard within minutes.
Track 2 — Drug Repositioning
-
1
Write your report
Prepare a written report proposing repositioned drug candidates supported by your analysis. Accepted formats: PDF or Markdown.
-
2
Submit before the deadline
Upload your report via the submission form. Late submissions cannot be accepted as reports go directly to human judges for scheduling.
-
3
Await judging
The independent panel reviews reports over a ~2–3 month window. Results are announced after the judging period closes.
FAQ
Who can participate?
Anyone with a Hugging Face account can participate. Dataset access is gated and approved manually — request access on the dataset page.
Can I participate in both tracks?
Yes. Track 1 and Track 2 are scored independently. You can submit to one or both.
How is Track 1 scored?
Submissions are evaluated on exact-match precision and recall across the four key fields: p_id, c_id, chr, and bp. Partial credit rules are pending final sign-off — check back for updates.
How many submissions can I make?
Only your best-scoring Track 1 submission is shown on the leaderboard. You may resubmit to improve your score. Track 2 accepts one final report per team.
What compute resources are available?
The challenge Space runs on CPU-basic hardware. You are welcome to use your own compute for training — only the final submission file needs to be uploaded here.
What is the data license?
The underlying dataset is gated and subject to the terms of the Hackathon Rules and Data Transfer Agreement — access requires approval, redistribution is prohibited, and all data must be deleted upon conclusion of the Hackathon. Participant submissions and results (predictions, code, reports) are released under CC BY 4.0 and may be reused with attribution.
When will winners be announced?
Track 1 rankings are visible live on the leaderboard. Track 2 results will be announced after the ~2–3 month judging window closes. The exact date will be posted in the HF Discussions page.
I have a question not answered here.
Post in the HF Discussions page. Organizers monitor it regularly.
Challenge Rules
Eligibility
- Open to individuals and teams worldwide. No prior affiliation with Sage Bionetworks or AWS is required.
- Participants must have a valid Hugging Face account and must request and be granted dataset access before accessing any data.
- Employees of Sage Bionetworks and AWS (and their immediate family members) are ineligible for prizes.
- Teams may consist of any number of members. Each individual may only be a member of one team.
Data Use & Privacy
- The dataset is gated. Access is granted only after completing the Data Transfer Agreement on the dataset page. By requesting access you agree to all terms therein.
- The data may be used solely for the purposes of this challenge. Redistribution, re-publication, or sharing of the raw data in any form is strictly prohibited.
- All downloaded data must be securely deleted upon conclusion of the hackathon or upon withdrawal from the challenge, whichever comes first.
- Attempts to re-identify the patient whose data underlies this challenge are prohibited and may result in immediate disqualification.
Submissions
- Track 1 submissions must be a CSV file following the schema defined on the "How to Submit" tab. Malformed files will be rejected.
- Track 2 reports must be submitted as PDF or Markdown. Each team may submit one final report.
- Late submissions cannot be accepted for Track 2 as reports go directly to human judges on a fixed schedule.
- Submissions must be the original work of the submitting team. Use of publicly available tools, libraries, and pre-trained models is permitted and encouraged, provided their use is disclosed.
- Organizers reserve the right to request code or methodology documentation to verify any submission.
Intellectual Property
- Participants retain ownership of their submissions.
- By submitting, participants grant Sage Bionetworks a non-exclusive, royalty-free license to reference and describe the submission in publications, reports, and communications related to this challenge.
- Participant submissions and results (predictions, code, reports) are released under CC BY 4.0 and may be reused with attribution.
Prizes & Judging
- Track 1 prizes are determined by the leaderboard ranking at the close of submissions.
- Track 2 prizes are determined by the independent judging panel using the criteria described on the Overview tab.
- Prize winners must provide documentation confirming eligibility before payment is issued. Prizes are awarded in USD. Recipients are responsible for any applicable taxes.
- Organizers' decisions on eligibility and judging are final.
Conduct
- Participants must treat all other participants, organizers, and patient data with respect.
- Attempts to manipulate the leaderboard, exploit scoring infrastructure, or interfere with other teams' submissions are grounds for immediate disqualification.
- Questions about rules or scoring may be directed to organizers via the HF Discussions page.