Code for our preprint is open-sourced!
🚀 Explore Frame2seq on GitHub:
github.com/dakpinaroglu...
📓 Try Frame2seq out in our Colab notebook:
colab.research.google.com/github/dakpi...
💻 Also try it out with:
`pip install frame2seq`
🧵 Work with Kosuke Seki, Amy Guo, Eleanor Zhu, Mark Kelly, and Tanja Kortemme
🧵 Our "Top0" design (0% seq. id. to Top7) is soluble, monomeric, folded, and the HSQC spectrum reveals that "Top0" likely maintains a fold identical to Top7!
🧵 We design and test novel proteins for de novo backbones. Most significantly, we evaluate a 0% sequence identity design for the famous Top7 fold
🧵 Frame2seq is an IPA-based encoder-only model that achieves state-of-the-art native sequence recovery
🧵 Super excited to share our model Frame2seq and our preprint “Structure-conditioned masked language models for protein sequence design generalize beyond the native sequence space”
www.biorxiv.org/content/10.1...
