Leash Bio

Very proud to announce our Dataset v2.0! When we first set out to build a universal data format, I wasn't sure if it would ever work. There is still work to be done, but by building on top of Zarr (https://lnkd.in/ee7vukjf) we're standing on the shoulders of giants. Come build with us! https://lnkd.in/eNA8fPhX And a big shout-out to Leash Bio, especially Andrew Blevins and Ian Quigley, PhD, for their trust and patience! With 100M+ rows and 120GB+, their awesome BELKA dataset was the perfect resource to test our assumptions and implementation. Check it out on Polaris now: https://lnkd.in/eFg3JyEN

BELKA, a massive public molecule-protein interaction dataset created by Leash Bio, is now available on Polaris! Access it with just a few lines of code: https://lnkd.in/gfY2ZYvs BELKA contains 100M molecules screened against 3 protein targets. If you didn’t get to participate in the Kaggle competition or just want to explore the dataset in its entirety, now’s your chance! 🚀

We just released 4.25B physical small molecule-chemistry measurements to the community on Polaris - Benchmarks for methods that matter, the biggest such public collection on earth. Please use it! Blog post here https://lnkd.in/g99A9YzC Dataset here https://lnkd.in/gpZtifhW More technical details here https://lnkd.in/gNPu8Uzj

We're thrilled to join the #AWSGenAIAccelerator2024 cohort. Thanks to Amazon Web Services (AWS) for this opportunity! https://lnkd.in/gczjjK4R

Amazon just announced the cohort of companies in its Generative AI Accelerator Program, and Leash is one of them! Blog post on why we're excited about it: https://lnkd.in/gV7YFuX7?

Thanks @dereklowe for the writeup on BELKA! https://lnkd.in/dgWNHr9c

NEW: Results from a protein binding contest suggest AI models still have ways to go in solving some of the more complex biological questions. I talked to Leash Bio CEO Ian Quigley, PhD, whose startup ran the Kaggle contest over the past few months, about how ~2000 competing teams fared: “They are less good, and I mean pretty poor but still doable, at predicting molecules that look different from the others but share a very common central core,” Quigley told me. “What the computers cannot do at this current stage is predict on molecules that don’t look anything like the ones they’ve been shown already.” “They are pretty good at memorizing and pretty bad at extrapolating into novel chemical space,” he added. “When I say pretty bad, you could randomly choose molecules and ask if they are binders or not from a collection, and you would be doing as well as the winning models in this competition.” A lot of interesting implications to consider from the so-so results: AI hype, generalization capabilities of models, capturing the right biological complexity in datasets, proper benchmarks for the field and more. Leash's big bet is that generating far more data will make better AI predictions. Quigley compared to to playing chess or folding proteins — ML problems that were only cracked when there were massive databases to train on. My latest for Endpoints News:

Our Kaggle competition to predict small molecule binding is over, and we're here to tell you that nobody was able to generalize into novel chemical space whatsoever. There's a lot of work to do! Blog post: https://lnkd.in/gW7VNaEa Endpoints: https://lnkd.in/gUefHgJg Kaggle competition: https://lnkd.in/gqcYrPnS

I'm excited to announce that Sarah Hugo has joined the team at Leash Bio as our Senior Research Associate! 🎉 She brings with her extensive experience at the University of Utah and Recursion. Her track record generating critical datasets, optimizing process, and just all around incredible skill in the lab make her an incredible addition! Welcome Sarah Hugo! So glad you're part of the team!