We are so excited to share the formation of MLBio! ππ
Looking forward to seeing what comes of merging two very vibrant scientific communities π§¬π¬
What a fantastic effort to do this alongside @ebetica@pkoo562 and @david_a_knowles
Sign up or follow @workshopmlbio to stay up to date!
@kdpisda@HWaymentSteele However we did show in the SI that using correct input structure for SLAE-CS increases the prediction accuracy a lot for ubiquitin alternative conformation compared with WT.
@kdpisda@HWaymentSteele Hi! We didnβt systematically answer this question because it requires the correct sampling of dynamics which itself is a bottleneck in the field.
@MagicInByte That's a good point! We didn't test the fine-tuned models on their original tasks (SLAE or ProteinMPNN). But considering that SLAE is trained using Rosetta score as one of targets, this may work!
1/ π§΅ New preprint! Do protein foundation models really understand local chemistry?
We probed them with NMR chemical shifts, per-atom reporters of chemical environment.
Short answer: they struggle exactly where the biology gets interesting π
[link] https://t.co/2DhcjYGM99
@LindorffLarsen@ginaelnesr@_YilinChen_@HWaymentSteele Thanks! The current QM methods seem to be even worse than both ML based or feature based methods for chemical shift prediction which may be due to the quality of input structure (ProCS15, AFNMR, etc.).
8/ Takeaways:
β Chemical shifts are a sharp, per-atom probe of what protein models learn
β Cleaner or more data won't close the gap
β Dynamics + atypical environments are a shared blind spot
β Models can learn to flag their own errors
@ginaelnesr and I are excited to release makeshift -- a lightweight package to lower the barrier for using NMR dynamics data in ML. We hope this can pave the way for more NMR-based models like Dyna-1!
Preprint: https://t.co/QsiqPtNirk
Docs: https://t.co/AVBTNgnpp9
Iβm excited to share that Iβll be in San Diego to attend the MLSB Workshop on December 7th! Iβll be presenting a poster titled βAn End-to-End Differentiable Framework for Protein NMR Assignment.β
Looking forward to sharing our work and connecting with everyone at MLSB!
β¨ Excited to share my new preprint: "Extending the MMPBSA method to membrane proteins: Addressing conformational changes upon ligand binding to P2Y12R" | ChemRxiv - https://t.co/ATKwVjEzYL