Pinned Tweet
We’re excited to share that BenchMHC, our open-source Python framework to reproduce, retrain, and evaluate peptide-MHC presentation models, is now available on @github! 🚀
🧵⬇️
Replying to @instadeepai @github
With BenchMHC, we've reproduced NetMHCpan 4.1 and can now directly compare retrained models against the original tool to understand what drives performance. 🎯
Explore BenchMHC today.👇
💻 Check out the open source repository, now on GitHub: bit.ly/4AYIjXx
Replying to @instadeepai @github
Goal: a clean benchmarking setup to separate model design from data and training effects. 🔍
We look forward to seeing how the life sciences community uses BenchMHC! 👏
Thank you all for a great @bitsinbio Paris!
On Tuesday this week, 70 of us gathered at our Paris office for three talks on proteins, cells and patients with @orakldotbio and @ScientaLab. 🇫🇷
Thanks to our speakers, and to everyone who joined us for the networking! 💪
Join @bitsinbio Paris, 22 Sept at the InstaDeep office! 🇫🇷
With State-space models for protein–protein interactions (InstaDeep), noise ceilings in cancer drug response prediction (@orakldotbio), and patient representations at scale (@ScientaLab).
Doors 6:30 PM, talks at 7, networking until 9.
🔗 Register your place: luma.com/0ddc69mb
🤝 Introducing STIX, InstaDeep's new JAX library for generative modelling, built on the stochastic interpolants framework.
STIX unifies flow matching, diffusion (including uniform and masked variants), and Bayesian flow networks under a single, common design space. 🧵
🧬 STIX handles continuous and discrete data as a single unit, generating both jointly. Discrete variables can be embedded in continuous space and run through the continuous interpolant, or handled directly via discrete interpolants like masked and uniform diffusion.
🚀 We're excited to see how you use STIX. Try it out today!
🐙 GitHub: bit.ly/4j2hZoD
📖 An introduction to STIX: bit.ly/3T69gqR
📚 Documentation: bit.ly/46iz0n7
🎙️ Calling our shots. In episode 3 of the Let's Talk Research podcast, Senior Research Engineer Christoph Brunken and Staff Research Scientist Jules Tilly explored our future ambitions for machine-learned interatomic potentials (MLIPs).
🧵⬇️
Since then, we've hit many of those milestones, with the inclusion of electrostatics in the library, advancements in reactivity, speed, and with more exciting news to be shared here soon.
📚 Hear our plans for what's next for the open-source mlip Python library and dive in yourself via the links below. 👇
🎧 Spotify: bit.ly/3JV6hgi
👀 YouTube: bit.ly/4p1KToz
🍎 Apple Podcasts: apple.co/4rdGnWr
⚛️ The library: bit.ly/47BX45J
Doors 6:30 PM, talks at 7, networking until 9.
🔗 Register your place: luma.com/0ddc69mb
"It’s an active rebellion against throwaway culture to try to preserve things"
📖 Mitch Johns, This Museum is (Not) Obsolete @Hack_Modular
Hear from author Jessica Holland (@hollandjess) as she follows the growth of PCB design through the musical history of vintage synths. 🎹
Read the full article in Decisive Agents 03: The PCB Issue: bit.ly/4ixCmcY
“Creating (a PCB) is a mixture of legacy knowledge, absolute understanding of electronics, and an eye for detail.”
Hear from veteran PCB engineer Steve Gascoigne of Siemens EDA (@siemenssoftware) as he explores the history and future of PCB design.
Read Jake Lourie’s [tag] interview in Decisive Agents 03: The PCB Issue. 📖
Get access: bit.ly/4x8Whnj




