Polaris - Benchmarks for methods that matter

If curating drug discovery-related data bores you, try Auroris. We’re making it simple, easy, and potentially… fun 😅 All it takes is a few lines of code to define the curation workflow: https://lnkd.in/g5Ayqj78

To help promote transparency reproducibility in data curation, we’re introducing Polaris Recipes! We want to make the ingredients (data sources) and steps (curation & preprocessing workflows) for a dataset on Polaris more visible. Get clarity on the data you're using: https://lnkd.in/g2cXt7gB Today, it’s simply a collection of notebooks outlining curation details. Inspired by Condaforge, we’re planning more robust ways of establishing transparency reproducibility on Polaris. We're also open to ideas & feedback!

If you’re working at the intersection of ML and drug discovery, you’ve got enough problems on your plate. That’s why we want to make it as simple as possible for you to access data on Polaris through a clean and intuitive API. Here’s Cas Wognum (again 🥹), sharing more about how we are hard at work to support any ML-ready dataset: https://lnkd.in/g_UsMM2K README? Read this👇 it’s all you need to load a benchmark, get the train and test data-loaders, evaluate predictions, and submit your results on Polaris. More tutorials here: https://lnkd.in/gHhqfgFU

Here’s Cas Wognum talking a bit more about Polaris and our roadmap at MoML 2024. https://lnkd.in/ggTpKSWU ⚙️ Currently in the works: hosting competitions, a review process for certifying datasets/benchmarks, guidelines (dataset curation, model evaluation & comparison), and more!

Let’s talk about data curation. Fun? 🤷 Important? Yes 🫡 High-quality, well-curated datasets are the foundation which we can build realistic, impactful benchmarks for drug discovery. We're excited to release Auroris, a library that simplifies the curation process for drug discovery datasets. Get started here: https://lnkd.in/g5Ayqj78 We’re starting with a focus on small molecules. Auroris currently has tools for: 🗄️ Molecule standardization 🔍 Detection of duplicate molecules with contradictory labels, activity cliffs between stereoisomers, and outliers 📽️ Visualization of outliers and molecular distributions To help with transparency and reproducibility Auroris automatically generates reports summarizing changes to a dataset during curation. The API makes it easy to define complex curation workflows that are serializable so you can share, review, or reproduce them!

Thank you to everyone who came out to the Polaris launch event! 🥳 We’re giving out merch boxes for everyone that uploads a dataset or benchmark before August 16th. Also, if you post your pictures from the photobooth and tag Polaris - Benchmarks for methods that matter, we’ll ship you some socks! Check out Polaris here: polarishub.io We’re excited to see how we can help host your datasets, benchmarks, or competitions on Polaris. If you’re interested in collaborating, drop us a note on our GitHub. https://lnkd.in/eZf6nnUM

🚀 Today, we’re excited to launch Polaris! 🚀 Polaris is a platform where the ML community working on drug discovery problems can easily share and access datasets and benchmarks. Get started with just a few lines of code: https://polarishub.io/ Benchmarking in computational drug discovery is already complicated because we don’t have standardized datasets, clear guidelines, or tools for method evaluation and comparison. It also doesn’t help that datasets and benchmarks are also scattered throughout the literature. We believe the first step towards improving the state of benchmarking is by providing a single source of truth for the community to easily access and compare methods. That’s what Polaris aims to be. ⭐ Whether you’re working with small molecules, phenomics, proteins, or something else, we’re here for you. We aim to support all file types and sizes with simple, fast data access. We’ve also put together a steering committee of experts across industry and academia (AstraZeneca, Relay Therapeutics, Pfizer, Merck, Nimbus Therapeutics, Blueprint Medicines, Johnson & Johnson, Valence Labs, and Mila - Quebec Artificial Intelligence Institute) to propose guidelines around dataset curation, method evaluation, and comparison - starting with a focus on small molecules. This is just the beginning of more industry-academia collaborations. Stay tuned!

If you are at ICML next week and into drug discovery, it might be very much worth it to check out the @Polaris_HQ evening event, I suspect :) I will certainly be there as well - would be nice to chat there. Here is their sign-up link: https://lu.ma/wj1agv8o

Today's benchmarking landscape for ML in drug discovery is complicated. We want to help the community develop methods that matter!🌟 Join us for our launch party at ICML on July 25th to learn more: https://lu.ma/wj1agv8o