How do you get other coordinates of amino ?

[HB-pencil-zero]

You only predict four coordinates in a residue, so how do you get other coordinates in a residue?

[kxz18]

The common way is to first use Rosetta to add and pack the sidechains, then optimize the energy with softwares of molecular dynamics (e.g. OpenMM).

[HB-pencil-zero]

I try to find the Internal context encoder code by searching 'internal', but I did not find the code. Can you tell me where is the code of Internal context encoder and External attentive encoder ?

[kxz18]

Codes for creating these layers:
internal context encoder: https://github.com/THUNLP-MT/MEAN/blob/main/models/MCAttGNN/mc_egnn.py#L273
external attentive encoder: https://github.com/THUNLP-MT/MEAN/blob/main/models/MCAttGNN/mc_egnn.py#L278

[HB-pencil-zero]

Hello author, this amino acid residue seems a bit weird, why does the carboxyl group only lose one H atom, it should lose a whole hydroxyl group in the condensation reaction?
.What's more , I’m curious about the predicted Z coordinate in what coordinate system, if it is in the absolute coordinate system, will the rotating antibody predict the same coordinate?

[kxz18]

Yes, this figure is wrong. We have corrected it in the camera ready version but the arxiv preprint has not been updated. It will be updated soon. The predicted coordinates are in absolute Cartesian system. If the input antibody rotates, the output will have the same rotation.