About BatV2Coords

[Nini-0120]

I found that 'root_3_xyz' has been used directly when converting vBAT back to Cartesian coordinates. Do the 'root_3_xyz's same as heavy atoms in residues? Does it mean the heavy atoms are fixed in reconstruction?

[truzmeto]

Hello Nini-0120, The 'root_3_xyz' variable holds the positions of three atoms selected from the N-terminal of the chain, which are referred to as "root atoms." These atoms serve as an anchor, carrying both translational and rotational degrees of freedom. During reconstruction or prediction, the positions of these three atoms are fixed, while the rest of the chain is allowed to move. The deep learning model learns the dynamics of these motions using vBAT.

When reconstructing known conformations (e.g., from training or validation sets), the positions of the root atoms across all frames can be processed in a batch. However, when generating new conformations (e.g., through interpolation or generative sampling), the root atom positions are borrowed from a reference conformation. Please let me know if you have any other questions.

[Nini-0120]

Thanks for the reply. I'm still confused that if the reference conformation contains only one frame, is that means every new conformations generated by model are sharing the same one 'root_3_xyz'?

[truzmeto]

Correct. Let me clarify. If we want to generate 1000 novel conformations, we simply make 1000 copies of 'root_3_xyz' from reference conformation and use it to initiate transformation from internal to XYZ. This means all generated conformations would have same reference. Attaching link to a paper for more details: https://www.biorxiv.org/content/10.1101/2024.05.05.592587v1