Contribute models

This tutorial explains how to incorporate models in the Ersilia Model Hub

This page serves as a guideline to develop a selected model to add to the Ersilia Model Hub. As described earlier, models can be of three types, namely:

1. Models developed by third parties.

2. Models developed by Ersilia based on publicly available data.

3. Models developed by Ersilia based on data from collaborators.

In this page, we focus on the first type of models, i.e. models developed by third parties. These third party models come with source code and parameters (or data to train them). We show how to use the Ersilia Model Template to adapt the assets provided by third parties and incorporate them into the Ersilia Model Hub.

Model List

First, check our Model Hub Backend, which contains:

• Model Hub: models that are already in the Hub or being incorporated at the moment. Each model should contain accurate information about its usage, source and any other relevant information.

• Model Suggestions: Contains a backlog of models that could be of interest to Ersilia with minimal information about them. There is a Selected tickbox that means that someone is already working on the model. Once a model is incorporated, it is filtered out of this list. If you have a new model suggestion, please fill in this form with as much detail as possible.

To start contributing:

1. We only accept contributions through GitHub. Please make sure you have an account and are familiar with the basic git commands (clone, pull and push)

2. Become a member of the Ersilia Slack Workspace. All our communications are centralised there. If you need a sign up link, please email us at [e-mail]

3. Join the #model-contributors channel in Slack and ask for edit permits for the Spreadsheet.

4. Fill in models of interest in the Model Suggestions list (if you need some sources of inspiration, check below)

5. Select a model of interest from the approved backlog and open a Model Request issue in the Ersilia repository.

6. Wait for the issue to be approved. This will trigger an automatic Action that will create a new repository for the model.

7. Fork the repository.

8. Start coding! Head to the Model Incorporation Guidelines for more information.

Know where to find models and get inspiration

Unfortunately, there is no centralized resource to find AI/ML models for drug discovery online (reason why Ersilia exists). The following resources can help you, though:

• Scientific literature: use Google Scholar and type relevant keywords (e.g. antimicrobial drug discovery).
• Reputable scientific journals: identify scientific journals that tend to publish AI/ML research related to drug discovery (e.g. Chemical Science, Nature Machine Intelligence, etc.). Subscribe to their newsletter if you are interested.
• Code repositories (mainly GitHub): on a regular Google search, type github next to your keywords for the search.
• Browse other model hubs such as Hugging Face, TensorFlow Hub, PyTorch Hub, etc.
• Expore Papers With Code. It is a fantastic resource.
• Follow benchmarks, competitions and leaderboards: Kaggle, DREAM Challenges, Therapeutic Data Commons, etc.
• Use Twitter: follow accounts like @KevinKaichuang, @andrewwhite01, @EricTopol, @iamtrask and @janjensen. From Ersilia (@ersiliaio, @mduranfrigola, @TuronGemma) we tend to retweet relevant literature.
• Use the #literature channel in the Ersilia Slack workspace to get inspiration.

• Check the models backlog in the Ersilia Model Hub to avoid duplications and get inspiration too.