@DdelAlamo@amirshanehsaz One possible explanation: Boltz-1's diffusion module isn't trained with weighted loss on UNK reference conformers, so providing a reasonable residue identity +conformer scaffold assists final design. It may also help refine trunk representations, though we haven't verified. (2/2)
@DdelAlamo@amirshanehsaz Thanks for the question, Diego! I'm one of the authors. Short answer: we saw it helped during model development, though we don't have an ablation visualization to share. (1/2)
@MarkNeumannnn Have you found any solution for indefinite hangs? I have found that the hanging usually completes when uploading an artifact, but it does hang quite often (typical hangs take minutes, sometimes hours).
Feeling lost in a sea of protein design AI papers?
Wondering what the state of the art is? What methods to use? What problems to work on?
We just put out a preprint to make your life a little easier.
https://t.co/u7wX8DZ1mW
#MLSB is back! Beyond excited for the range of our invited speakers this year!
Bridget Carragher @bcarra2
Kyunghyun Cho @kchonyc
Rhiju Das @RDasLab
Polly Fordyce @fordycelab
Tanya Kortemme @KortemmeLab
RFDiffusion team
See you in December for biomolecules @NeurIPSConf !
Reflecting on AlphaFold 2. It's a strange system. It's not an explanation of protein structure, nor a theory of protein structure
And yet... there is a sense in which it (& successor systems) capture much (not all) of our best understanding of protein structure
As a part of the #llmhack, me and Sean Warren developed MolVerse, a language interface to 3dmol.js: a web app for all chemists and biologists that want to visualise their (bio)molecules without needing to code. Here we load the structure of benzene:
I’ll be sharing my recent work on all-atom protein structure prediction with SidechainNet at tonight’s COMP poster session 7-9pm at ACS. If you’re attending, come by! 10 (!!) posters from our research group will be on display covering deep learning and drug discovery. #ACS2023
Exciting news from The MolSSI! Our BEST PRACTICES workshop site, with a new user-friendly theme and dark mode, is now live thanks to the leadership of Dr. Jessica Nash @Jessica_A_Nash, MolSSI's Education Lead. Check it out at https://t.co/9DTjDmudlH
Eager mode was what made PyTorch successful. So why did we feel the need to depart from eager mode in PyTorch 2.0?
Answer: it's the damn hardware!
Let's tell a story about how the assumptions PyTorch were based off of became untrue, and why PyTorch needed to evolve. (1/10)
DiffDock is now on @huggingface Spaces with interactive visualization of the diffusion steps. Built using @Gradio and 3dmol.js and you can upload your own ligands as SMILES or sdf/mol2. Happy docking! https://t.co/nHiGYY6QWQ
1/2
We present Folding Diffusion: a diffusion model for protein structure inspired by physical protein folding.
Let by @Kevin_E_Wu during his internship, with @alexijielu@vdbergrianne@james_y_zou@avapamini
Code: https://t.co/ldKQGnsrMQ
Preprint: https://t.co/fGdHqLTi5U