Release first.
Publish after review.
A journal ambition does not automatically require a six-month model embargo. Keep the public release and peer-reviewed paper on separate schedules—and clarify publicity before launch.
Provisional advice: the model’s biological domain, validation results and proposed access model have not been supplied. Venue fit below is an assessment, not an acceptance prediction.
DeepSeek is a real precedent. Bio has its own.
| Model | Public milestone | Journal milestone | What this establishes |
|---|---|---|---|
| DeepSeek-R1 | Model: 20 Jan 2025 Preprint: 22 Jan 2025 | Nature: 17 Sep 2025Received 14 Feb; accepted 17 Jul 2025. | The model and manuscript were public before the journal received the submission. |
| Evo | Model, weights and research release: 27 Feb 2024 | Science publication announced: 14 Nov 2024 | A particularly relevant biological foundation-model example, with additional experimental validation developed after the preprint. |
| ESM3 | 25 Jun 2024: preprint and small open-model release | Science: Jan 2025 | Public research and an open checkpoint preceded publication. This was not release of every model size; the broader API beta came later. |
Precedents show compatibility in those cases, not advance approval of every future product launch. They also show why release-to-publication time should not be confused with time you must keep a model private.
Choose the scientific claim, then the journal.
The middle column is my assessment of what would make a persuasive submission, informed by each journal’s scope. It is not a universal checklist imposed by the publisher.
| Venue | Best fit / evidence to emphasize | Early release and submission route |
|---|---|---|
| NatureHighest-ambition general-science route | A capability or insight with consequences across biological research. Show what becomes scientifically possible, not only a higher benchmark score. Purely technical advances can qualify if their research impact is substantial. | Nature Portfolio preprint policy applies. Optional presubmission enquiry; usually answered within two working days, not guaranteed and not a promise of review. |
| ScienceHighest-ambition general-science route | A compelling biological foundation-model story with broad significance. Evo and ESM3 are close precedents; strong demonstrations of real biological function would strengthen the case. | Release-first precedents verified. The current author/policy pages blocked automated access; confirm present rules and the proposed launch plan with the editorial office. |
| Nature BiotechnologyStrong candidate for bioengineering / useful technology | Protein or molecular design, genomics, bioengineering or a broadly useful biotechnology platform. Emphasize the capability’s practical biological value and convincing validation. | Nature Portfolio preprint policy applies. Does not accept presubmission enquiries; prepare a full research submission. Policy questions are distinct from manuscript pitches. |
| Nature MethodsStrong candidate for a general-purpose research method | A tool for basic life-science research. Its scope explicitly asks for strong validation, application to an important biological question, and comparisons with existing methods. | Nature Portfolio preprint policy applies. Accepts abstract-based scope enquiries; submit the full manuscript directly if already written. |
| Nature Machine IntelligenceStrong candidate when the AI contribution is central | Novel learning methods or generalizable AI capabilities, with biology as a substantial application. Emphasize technical insight, ablations and evidence of generalization. | Nature Portfolio preprint policy applies. Does not accept presubmission enquiries; full manuscript route. |
| Nature CommunicationsBroader alternative, still selective | A substantial, well-supported advance important to the relevant specialists. A strong option if the evidence is convincing but the story is less compelling to general-science editors. | Nature Portfolio preprint policy applies. Use the submission guide. Do not assume a faster review merely because it is an alternative venue. |
| Science AdvancesAdditional multidisciplinary candidate | Significant original research spanning AI and biology. AAAS’s author brochure describes its broad scope; that brochure is historical, not a source for current fees or detailed rules. | Current author/policy pages could not be retrieved. Confirm preprint, launch/publicity and submission requirements directly before relying on this route. |
Working ranking: Nature or Science for a genuinely field-changing result; Nature Biotechnology or Nature Methods for a powerful biological platform; Nature Machine Intelligence if the main advance is in AI.
Wet-lab is not a universal formal requirement. My advice: match validation to the claim. If you claim new biological function, prospective experiments are much more persuasive than model-based predictions alone. For an analysis method, rigorous real-data evaluation and a meaningful biological application may be appropriate. Nature Methods’ scope requires validation, not one mandatory experimental format.
Preprints are not press campaigns.
What Nature explicitly permits
Research preprints do not count as prior publication and can be posted during review. Disclose the preprint identifier and licence when submitting. Conference proceedings are different: a later journal paper needs a substantial extension. Preprint / proceedings policy.
Weights and code releases have strong precedents, but those examples are not a blanket policy ruling on an API product, commercial white paper or launch event.
What needs care
Nature discourages actively soliciting prepublication media coverage. Responding to unsolicited questions about a preprint is permitted; make its unreviewed status clear. Accepted-paper publicity has separate embargo rules. Press policy.
Before interviews, a coordinated press launch or a commercial product announcement, describe the exact plan to the editor or press office and request written guidance.
Verification limit: Science’s core policy and author pages returned access errors. Science’s past publication of released bio models is independently documented above. That establishes precedent—not confirmation of its current detailed media or prior-publication wording. No Science Partner Journal policy has been substituted for Science’s own policy.
A scientific evidence package, not just a model card.
Verified Nature Portfolio obligations
Provide data- and code-availability statements, disclose access restrictions at submission, and supply central custom code to editors/referees when requested. Data needed to assess the claims must be available under acceptable conditions. Public reproducible code is encouraged; do not assume every paper must release every byte of pretraining data, or that a closed API automatically suffices. Reporting and availability policy.
Report relevant research approvals and human-data protections. Where misuse risks apply, complete the dual-use section of the reporting summary; additional specialist review may be required. Journal review does not replace a responsible release assessment. Research ethics policy.
My recommended AI-for-biology package
- One precise claim: what can researchers do now that they could not do before, and why does it matter?
- Credible evaluation: strong current baselines; independent test data; checks for train–test overlap; ablations separating architecture, data and scale; uncertainty and failure analysis.
- Biological evidence: a meaningful application and validation proportionate to the claimed capability. Do not present predicted function as experimentally established function.
- Reproducibility: identify the exact evaluated checkpoint, data versions, preprocessing, training recipe, evaluation scripts, dependencies and compute. Preserve a stable release separate from ongoing development.
- Access plan: state what is public, gated, available to reviewers, licensed from others, or unavailable—and why. Agree any restrictions with the journal early.
- Manuscript: a full research article with methods, limitations, supporting results, references, author contributions and disclosures. Use the target’s current article instructions rather than treating a technical blog as a ready journal submission.
For Nature, see the initial-submission checklist and formatting guide. Exact lengths and fees are venue/article-specific and are not estimated here.
Keep the release date. Make the paper stronger.
- Before launch: decide the main scientific claim, audit supporting evidence, clear release rights and safety considerations, and prepare a complete preprint plus a versioned model/evaluation package.
- Clarify editorial fit and publicity: Nature and Nature Methods offer enquiry routes. For Nature Biotechnology and Nature Machine Intelligence, use full submissions for scientific assessment; send policy questions separately. Disclose the intended date, weights/API availability, technical blog and any press activity.
- Release: post the research preprint and the agreed model package. Describe the findings as unreviewed, not as accepted or endorsed by a journal. Do not promise clinical utility from preliminary evidence.
- Review in parallel: submit to one journal at a time. Add requested validation and update the research record appropriately while maintaining access to the originally evaluated version.
- If declined: move to the next appropriate journal; a completed public release need not be reversed. Transfer or reframe the paper rather than leaving the product waiting for a decision.
There is no reliable six-month acceptance guarantee. Nature says an early decision not to consider a paper usually arrives within a week; this is not a peer-review or acceptance turnaround. DeepSeek’s receipt-to-publication interval was about seven months, despite the model already being public. Nature’s process; DeepSeek article history.
A short policy enquiry to adapt
We are preparing a manuscript describing [specific scientific advance], supported by [key validation]. We intend to release a research preprint on [server/date], together with [weights, inference code, evaluation data and/or API access]. Our communication plan includes [technical blog, social posts, interviews or press release].
Would these specific activities affect eligibility for consideration or your publicity policy? Are any proposed access restrictions incompatible with review or publication? We would appreciate written guidance before launch.
This asks about policy; it does not request scientific assessment from a journal that does not accept presubmission enquiries. No editors have been contacted on your behalf.