Submit your model to the platform
The Platform aims to provide biologically impactful AI models that help scientists explore the molecular underpinnings of human health and disease, with the broader goal of accelerating science and advancing biomedical research. The Platform thrives on community contributions. By sharing your models on the Platform, you make your work more discoverable, help advance AI in Biology, and enable broader research. In this guide, you’ll learn everything you need to know about submitting your models to the Platform.
Why Submit Your Model?
Showcase your model to an audience of researchers, developers, and computational biologists in the AI x Biology community.
Highlight that your work is accessible and easily reproducible by others.
Contribute to a growing library of models that drive innovation and collaboration.
See how your model performs against an expanding suite of standardized benchmarks.
Who Can Submit a Model?
Anyone can submit a model for consideration to the Platform.
If you’re a model developer, follow the submission guidelines below.
If you would like to suggest a model that you did not develop to be included on the platform, and/or benchmarked, please fill out this form.
Model Submission Guidelines
CZI is the final decision maker on which models will be accepted onto the platform. We are currently accepting model submissions in the following domains:
Single-cell transcriptomics
Perturbation prediction
Variant effect prediction
Imaging models, including:
Image Feature Extraction
Virtual Staining
CryoET
If your model falls outside these areas but is relevant to virtual cell modeling, CZI’s Scientific Grand Challenges, or biological research more broadly, we encourage you to reach out at virtualcellmodels@chanzuckerberg.com. We’re always interested in learning about emerging use cases that may align with the Platform’s mission.
After submission, our team will evaluate it against the criteria listed below. Meeting these standards helps ensure your model can be successfully integrated and provide maximum value to the community.
The model offers significant value to researchers and developers, and must include a preprint or publication. This could include:
Novelty - the model offers a new approach, solves an unsolved problem, or provides a unique capability.
Utility - the model addresses a clear need within the AI x Biology domain related to the Platform’s mission.
Performance - the model demonstrates strong performance on biologically relevant tasks.
Accessibility - the model is well-documented and easy for others to understand and use.
The model was developed with responsible AI principles in mind.
The model is fit for its intended use within a research context.
The model is available under an appropriately permissive license.
We want to ensure that the models hosted on the Platform are accessible and easy to use by developers and biologists. To support this, we require or strongly encourage each model be submitted with the following assets:
Item |
Description |
Use |
|---|---|---|
Model Card (required) |
A summary of the model’s purpose, training details, and biological relevance. |
Helps users understand the model’s intended use, limitations, and evaluation results, which supports reproducibility, responsible use, and trust. |
Quickstart Guide (required) |
Step-by-step instructions, formatted in a notebook, for running the model. See template here. |
Lowers the barrier to entry for others to use the model, increases model adoption, and helps ensure the model is used correctly with minimal setup. |
Tutorial (strongly encouraged) |
A hands-on, guided example demonstrating real-world use or retraining of the model. |
Increases adoption, especially by a biological audience, by providing a guided example that demonstrates a real-world application. |
(Future) Wrap your pip-installable model with MLflow to enable standardized packaging and submission. |
Standardized interface for wrapping and executing your model. |
Simplify integration with future downstream applications (such as benchmarking and other applications) and enable use of the model by developers and biologists. |
How to Submit a Model
If you would like to submit your model, please email us at virtualcellmodels@chanzuckerberg.com, subject line “Model Submission: [NAME OF MODEL]” .
In your email, please describe how your model meets the criteria listed above and attach a PDF of the preprint or publication associated with your model. Work in progress manuscripts will be kept confidential to our internal teams.
When you submit a model, you agree that you have all necessary rights and permissions to provide the model, it does not infringe any third-party rights, and it complies with all applicable laws, our Terms, and our Acceptable Use Policy.
What Happens When You Submit a Model?
When you submit a model, it will go through a structured process to ensure quality and relevance. Here’s a general overview:
Initial Review: CZI staff will perform a check to ensure the submission is complete and meets our submission guidelines. If it does not meet these requirements, or is not aligned with our mission, we will send a rejection with the reasoning for the decision.
Content Review: Your model and its accompanying assets will be reviewed for clarity, accuracy, and adherence to our guidelines.
Security & Technical Testing: We reserve the right to perform additional security testing on submitted models and/or apply security controls as necessary to protect the integrity of the Platform and ensure its safe operation.
Benchmarking Consideration: We will assess if your model is suitable for inclusion in our standardized benchmarking suite based on task fit. If your model is benchmarked, we will share the initial results with you via email for feedback. For models submitted by a third-party (not the model developer), we will also share the results with the developer. We are looking for feedback on: a. Reproducibility: Can the developer use cz-benchmarks to reproduce the results that we obtained? b. Accuracy: Does the implementation run the developer’s model in the correct way? c. Task fit: Is this an intended use of this model for this task?
Publication: If accepted, your model and its Model Card will be published on the Platform and available for download for others to use. If benchmarked, its performance results will also be displayed.
Review Outcomes & Timelines
Once you submit your model, here’s what you can expect:
You will receive an email acknowledging your submission.
Our team will review your submission and provide a decision within four weeks.
Publication (if accepted): If your model is accepted, you will receive notification and final review of submitted assets. It will typically be published on the VCP within two weeks following the review decision.
We aim to provide a reliable and timely response to all submissions, and you can expect the complete process from submission to publication to take up to 6 weeks. If you need to get your model on the platform quickly to coincide with an upcoming publication, please let us know and we will expedite the process.
Possible Outcomes
Accepted & Benchmarked: Your model will be published on the platform.
Accepted (Not Benchmarked): Your model will be published on the platform, but it may not be included in our current benchmarking suite. This could be due to factors like resource availability, alignment with current benchmarking priorities, or the model’s specific nature.
Not accepted: Your model does not meet our submission criteria. We will provide specific feedback on why it was declined, and you may be able to resubmit after addressing the feedback.
Resubmitting or Updating Your Model
If your model was rejected, you are welcome to resubmit it once you have addressed the feedback provided by our team.
If you update your model, you can submit a new version. We encourage clear versioning (e.g., v0.0.0 or YYYY-MM-DD).
Licensing Your Submission
If your model is accepted and published on the Platform, all descriptive text, metadata, and images you provide for your Model Card, Quickstart, and tutorials will be published under CC-BY.
We require your model’s code to be made available under a permissive open-source license (e.g., MIT, Apache 2.0, BSD 3-Clause). This ensures the reproducibility and broad usability of your model within the community. We may not be able to host models with highly restrictive licenses.