Publish with us

Let the work outlast the leaderboard

A leaderboard position lasts as long as the leaderboard. Prize money matters, but it is temporary. What can endure is a publication that shows what actually works when a model is asked to predict an embryo, written with the people who found out.

This Challenge is therefore about more than rank and prize. It is also an invitation to help build the scientific record.

The Challenge analysis paper

planned

We are preparing an analysis of the Virtual Embryo Challenge for submission to Nature Methods. Following a discussion with an editor there, the paper will examine what the tasks measured; which approaches worked, which did not, and why; where the metrics agreed with biology and where they came apart; and what the results tell us about predicting development.

Leading teams from both the Human Team and Agent Team tracks will generally be invited to contribute, subject to applicable U.S. laws and policies. We intend this to be substantive co-authorship, not courtesy credit. Contributors will describe their own methods, help analyse or interpret the results, review the manuscript, and take responsibility for their contributions in accordance with the journal’s authorship requirements.

This is a submission plan, not an acceptance or a guarantee of publication. Peer review belongs to the journal. What we can commit to is inviting eligible leading teams into the process, including qualifying contributors as co-authors, and making a serious, good-faith effort to submit the work and revise it in response to review.

Data permission is required

required

The Challenge dataset has not yet been published. Any use of it in a publication or public presentation requires prior written approval from Dr. Neil Chi and the data-generating team. Any resulting publication must also appropriately credit Dr. Qingquan Zhang and Dr. Neil Chi.

If you are considering a paper, contact the organisers early so that this approval can be coordinated before submission. Treat the written approval, not an invitation to the collection or an informal discussion, as the formal condition for using the data. See Terms of Use, clause 14.

What this asks of you

For the Challenge analysis paper, focus on competing well and documenting the method that produced your result. If you are invited, that record will become the starting point for your contribution to the manuscript.

If you are planning a paper of your own that uses anything from the Challenge, tell us early so we can coordinate timing and the data approval below. Your manuscript and its scientific argument remain your own. Both tracks are eligible.

As you work:

Keep a running method summary. It is already a condition of receiving a prize, and it is the first draft of your contribution to the analysis.
Disclose external data as you use it. Record the source, version, licence, stages or conditions used, and how it entered the method. A result that cannot be described cannot be analysed or published responsibly.
If you are in the Agent Team track, preserve the complete trajectory. Record the LLM and version, prompts, harness, tools and permissions, budgets, evolutionary or search procedure, and code history. These materials are required evidence for the track and are the basis for analysing how far an agent can get, what enables progress, and where it stops.

You retain ownership of your method, code, and models. See the Challenge Rules, section 15.