Help understanding some NaN outputs

[AndreeaMusat]

Hi,

First of all, thanks for sharing this repo, your code is really useful!

I am trying to use MP-Nerf on some custom data. With Jonathan King's help (from SidechainNet), I managed to extract the necessary protein data for arbitrary PDB files and MP-Nerf works well for a large share of the proteins in the dataset. The dataset consists of proteins which have one chain and no gaps (we consider there is a gap if the distance between the C atom of a residue and the N atom of the next residue is larger than 1.7Å). Also, as sometimes there are some missing atoms, pdb2pqr was used to fix the PDB files.

However, for some proteins there are some NaN values and I cannot figure out why (or if further processing is needed for the PDB files in order for MP-Nerf to work well on those too).

I put the code here so that you can easily reproduce the issue on one of the problematic proteins.

Thank you!