Is PPMI (DUA-gated) acceptable as external data for prize eligibility?

DaT Parkinson's Prediction Challenge
1

Hi organizers,

The rules state that external data and pre-trained models are allowed, but that
for prize eligibility external data must be “freely and publicly available to all
participants” and licensed to permit commercial use (no NC / CC NC / CC BY-NC).

I’d like to confirm how this applies to PPMI (Parkinson’s Progression Markers
Initiative, ida.loni.usc.edu), which is the largest public DaT-SPECT collection
and the obvious external candidate for this task. Its access model is:

  • Free of charge and open to any researcher — no fee, no selection committee for
    the standard tiers.
  • But gated: you must register with LONI IDA and sign a Data Use Agreement
    before downloading.
  • The DUA restricts redistribution (data may not be shared beyond the
    investigators named in the request), limits use to the purpose described in the
    application, and attaches IP policies.

So PPMI is “freely available to anyone who applies”, but it is not downloadable
without an individual agreement and it is not released under an open license in
the CC/MIT sense. That seems to sit right on the boundary of the rule, hence the
question.

Specifically:

  1. Does PPMI satisfy “freely and publicly available to all participants” for
    prize eligibility, given the registration + DUA gate?
  2. Does the DUA’s purpose-limitation and no-redistribution clause conflict with
    the commercial-use requirement?
  3. Does the same answer apply to weights pretrained on PPMI — either trained
    by me, or third-party released checkpoints? Is a model derived from DUA-gated
    data treated like the data itself?
  4. If PPMI is not eligible, is it still permitted for non-prize-eligible
    submissions or for methodological exploration, provided the final submitted
    model is trained only on challenge data?

The same question applies to other DaT datasets released under “academic use
only” terms, such as the NTUA Parkinson dataset.

Clarifying this early would help a lot with planning. Thanks!

2

Hello,

Great question! The DUA states that “any model that has been trained using participant-level data may only be shared through MJFF-approved methods,” and it grants access to the data “for the purpose of scientific investigation, teaching, or the planning of clinical research studies.”

Because competition use doesn’t fall under those purposes, and because winning models must be openly licensed for commercial use, models trained on this data wouldn’t be eligible for prizes.