You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hi, first of all thank you for open sourcing this amazing tool!
The paper mentions that
The final layer of the end-to-end version of our 3-track network generates 3D structure
models by combining features from discontinuous crops of the protein sequence (two segments
of the protein with a chain break between them). Because the network can seamlessly handle
chain breaks, it can be readily utilized to predict the structure of protein-protein complexes
directly from sequence information.
Looking at the code in Refine_module.py and predict_e2e.py, it looks like there are quite a lot of details to get right and many ways to screw up how the information from two disjoint chains should be merged. According to the supplementary material there's also changes in the positional encoding needed for this to work (the +200 offset).
Is there an example script or some guidelines that you could share that shows how to properly do a structure prediction for a Protein-Protein complex?
Finally, I was wondering if any experiments were done trying to estimate binding affinity scores from the hidden layer activations of the Refine_Module? This would be a very useful extension for doing protein design...
The text was updated successfully, but these errors were encountered:
@minkbaek I still have some questions about the complex model especially the 2track.
it seems that the 2track has only the npz output without the pdb?
This seems to be different from the paper?
Thanks!
Hi, first of all thank you for open sourcing this amazing tool!
The paper mentions that
Looking at the code in
Refine_module.py
andpredict_e2e.py
, it looks like there are quite a lot of details to get right and many ways to screw up how the information from two disjoint chains should be merged. According to the supplementary material there's alsochanges in the positional encoding
needed for this to work (the +200 offset).Is there an example script or some guidelines that you could share that shows how to properly do a structure prediction for a Protein-Protein complex?
Finally, I was wondering if any experiments were done trying to estimate binding affinity scores from the hidden layer activations of the Refine_Module? This would be a very useful extension for doing protein design...
The text was updated successfully, but these errors were encountered: