peptoneltd / nerfax Goto Github PK

While trying to understand how the internal coordinates are represented on in SCNet format, I noticed nerfax seems to assign additional data beyond the bond and torsion angles and the bond lengths.

I noticed this when I added the following lines of code to your PDB reconstruction example.

seq = list(traj.topology.to_fasta()[0])

glycine_angles = [internal_coords["angles_mask"][0][i,:] 
                  for i in range(len(seq)) if seq[i] == "G"]
glycine_torsions = [internal_coords["angles_mask"][1][i,:] 
                  for i in range(len(seq)) if seq[i] == "G"]
glycine_bonds = [internal_coords["bond_mask"][i,:] 
                  for i in range(len(seq)) if seq[i] == "G"]

Examining these values, I find that each has five non-zero entries, even though point cloud mask has only four, and the data format described in mp_nerf.kb_proteins suggests that glycine should have only four bond angles, four torsions, and four bond lengths.

Notably, I only see this for glycines.
Everything else has a number of non-zero entries matching the mp_nerf description and number of nonzero entries in the cloud mask. Furthermore, the first four bond angles tend to be very near their reference values, but the last one varies greatly. Also, the final bond length is typically very large.

I have not tried out mp_nerf to see if this also happens in their code.

Is this intentional? If so, what are the extra bond lengths, torsions, and angles?

Thanks for the excellent package! To clarify, does it work with RNA PDB structures?

Hi, is there any example how to use nerfax with PDB files? Basically, how to use nerfax to convert 3D coordinates to internal coordinates and use internal coordinates to reconstruct 3D coordinates? Thanks!