search

lamoureux-lab / TorchProteinLibrary

PyTorch library of layers acting on protein representations
https://lamoureux-lab.github.io/TorchProteinLibrary/
MIT License
116 stars 23 forks source link

Beta angles appear to be incorrect #35

Open mpdprot opened 3 years ago

mpdprot commented 3 years ago

I reconstructed the backbone for the protein 6MSP from the PDB, using real dihedral angles fed to Angles2Backbone.

I calculated planar angles for both the original backbone versus for the reconstruction.

The planar angles N-Cα-C etc. are constant in the reconstruction. In the original, they are of course slightly variable. However it looks like the typical values used in Angles2Backbone do not correspond to the correct triples. It is currently: β(3j) = π − 2.1186 β(3j + 1) = π − 1.9391 β(3j + 2) = π − 2.061

However, from looking at this data, it looks like this is supposed to be: β(3j) = π − 2.0610 β(3j + 1) = π − 2.1186 β(3j + 2) = π − 1.9391

image
lupoglaz commented 3 years ago

Thank you, checking it right now.

truzmeto commented 1 year ago

The same problem is there for FullAtomModel.Angles2Coords, can you please look into it? Thanks!

truzmeto commented 1 year ago

Fixed it for FullAtomModel.Angles2Coords here: https://github.com/truzmeto/TorchProteinLibrary/blob/9d08825248e362cac8888bca432f87e99fb88314/Layers/FullAtomModel/cGeometry.cpp#L17

Tested reconstruction with 42 residue peptide, which gave 1.6AA backbone rmsd difference. Previously difference was ~7AA when beta angle mean positions were swapped. download