🧪 We're running a cheminformatics notebook competition!

Enter by October 4
Marimo Logo
Star
22.5k
GitHub

Bring Cheminformatics to Life: molab Notebook Competition #3

OpenADMET x marimo, co-hosted with Pat Walters

We’re challenging you to do what OpenADMET and marimo both believe in: turn real drug-discovery results into something tangible, reproducible, and interactive — that is, a marimo notebook.

In our first cheminformatics subcommunity molab Notebook Competition, you’ll dig into real ADMET (Absorption, Distribution, Metabolism, Excretion, Toxicity) data by building a marimo notebook around it. Pick from a curated set of OpenADMET datasets (originating from OpenADMET’s own blind-challenge datasets and baseline models) or propose your own RDKit-based idea.

We don’t expect you to beat OpenADMET’s leaderboard or fully reproduce a model. Instead, hone in on one idea — a dataset, a technique, a liability like CYP inhibition — and bring it to life with code, UI elements, and explanatory text. One aspect of this competition will be creativity, whether that is a novel visualization, an extension of a baseline model, or applying a technique to a new dataset. Anyone who works through your notebook should walk away with real intuition for the concept, and a sense of how to try it themselves.

For an example of the kind of submission we’re looking for, check out the submissions from our previous notebook competition winners. To learn what is possible with marimo, see our gallery.

You will submit your notebooks as a molab link. Because molab offers free GPUs, your submission can also run on a GPU.

Resources

  • marimo-chem-utils — Pat Walters’ open-source package for RDKit-native rendering inside marimo (molecule images, linked SAR tables, scatterplot-with-structure tooltips, structural-alert/filter review). You’re welcome to extend it, not just use it as-is. This repo also has links in the README to run the three demos on molab.
  • “Practical Cheminformatics with Marimo” | practical_cheminformatics_tutorials repo — Pat’s own walkthrough of building cheminformatics notebooks in marimo, with example notebooks already live on molab.

OpenADMET datasets

Examples from the OpenADMET datasets and models page include:

  • OpenADMET’s live CYP Inhibition Blind Challenge (announcement) — CYP3A4/2C9/2D6/1A2 inhibition and time-dependent inhibition (TDI). Build your notebook using the released datasets — think an explainer of what CYP inhibition is and why it matters, or an exploration of the training data’s structure-activity patterns. (Note: this notebook competition is completely separate from OpenADMET’s competition.)
  • PXR Challenge dataset (28,700 rows) — experimental dataset for predicting human Pregnane-X Receptor (PXR) induction, comprising over 11,000 compounds screened using a high-fidelity in-house assay.
  • ExpansionRx Challenge data (9 ADMET endpoints — logD, solubility, microsomal stability, permeability, plasma protein binding, etc. — CC-BY-4.0, 15,200 rows) — real-world ADMET data from a recently prosecuted series of drug discovery campaigns by Expansion Therapeutics on RNA-mediated diseases.
  • ASAP-Polaris-OpenADMET antiviral dataset — potency, ADMET and crystallography data from ASAP Discovery’s SARS-CoV-2 / MERS-CoV main protease inhibitor program.
  • Octant CYP Inhibition & Reactivity dataset (51,400 rows, CYP3A4/CYP2J2) — self-consistent, multi-endpoint CYP data generated on a single platform under controlled conditions, with full well-level readouts and quality annotations.
  • Wildcard track: propose your own RDKit-based idea if you’d rather chart your own course.

Who you are

Cheminformatics practitioners, computational chemists, grad students, RDKit contributors, and ML-in-drug-discovery folks. Research-curious newcomers and veterans alike.

How to enter

  1. Pick from a selection of datasets and models from the OpenADMET community.
  2. Build a marimo notebook that answers a question about the data.
    • We’re not looking for a summary of the dataset, but more so a question you actually want the answer to.
    • A few different example directions: how do the chemical structures in the test set differ from the training set? What happens to a model’s accuracy when you change how the data is split? Can you type in a molecule and see predictions across all four CYP isoforms at once? How many compounds survive when you enforce every ADMET threshold at the same time? Where do the published baseline models fail, and is it their fault or the data’s?
    • We are not judging on model accuracy, but more so on what the notebook reveals and whether a stranger can learn something from it. Notebooks that show us something we hadn’t thought about will beat notebooks that do the obvious thing well.
  3. Submit it by sending us the molab link and a video explainer through the form below.

Notebooks that are unrunnable will automatically be disqualified.

Judging criteria

Submissions are reviewed by a panel that includes Pat Walters and 2 members of the marimo team. The jury favors:

  • Notebooks that provide an intuitive understanding through code, UI elements, and text
  • Chemical validity
  • Creativity and UI innovation
  • Notebooks that implement custom extensions that either improve on a baseline or surface new insight

The full rubric is linked here for your reference.

Note: We grade each submission by hand. The rubric rewards customization (e.g. custom widgets, custom packages) because seeing the same AI-prompted notebook gets a bit old by the 20th one. We’re not discouraging AI use at all. Disclose it in the notebook (like Jesse Hartman’s example) and make the notebook actually yours.

Prizes

You’ll walk away with real, hands-on fluency in an ADMET concept. We’re also offering $2.5K+ in prizes.

  1. 🥇 1st place: Mac Mini, co-authored blog post, a marimo swag bag, + shoutouts on Pat Walters’ & marimo socials
  2. 🥈 2nd place: Claude Max Pro for 3 months, co-authored blog post, a marimo swag bag, + shoutouts on Pat Walters’ & marimo socials
  3. 🥉 3rd place: Claude Max 5x for 3 months, co-authored blog post, a marimo swag bag, + shoutouts on Pat Walters’ & marimo socials

Submit

Ready to build? Submit your molab notebook before October 4, 2026, 11:59 PM PST.

Submit here!

Stay in the loop

Follow along for discussions, submission highlights, and winner announcements:

Winners event

We’re hosting a joint virtual event with both Pat Walters’ and marimo teams on October 13, 2026 to reveal the winners. Everyone is welcome to join us virtually.

Register for Event