martinize2 AlphaFold benchmark

[Marco447]

I recently processed 200.000 structures of the AlphaFold database (martinize2 conversion and energy minimization), where I encountered 6 errors (all unknown modifications). I listed them below:

AF-O80995-F1-model_v3.pdb
19012
    INFO - general - Read 1 molecules from PDB file clean.pdb
    INFO - step - Guessing the bonds.
    INFO - general - 1 molecules after guessing bonds
    INFO - step - Repairing the graph.
    INFO - general - Applying modification N-ter to residue A-MET1
    INFO - step - Dealing with modifications.
    INFO - general - Identified the modifications ['N-ter'] on residues ['MET1', 'MET1', 'MET1', 'MET1']
 WARNING - unknown-input - Could not identify the modifications for residues ['CYS80'], involving atoms ['597-OXT']
    INFO - step - Read input.
    INFO - step - Creating the graph at the target resolution.
    INFO - general - Applying modification mapping ('N-ter',)
    INFO - step - Averaging the coordinates.
    INFO - step - Applying the links.
    INFO - step - Placing the charge dummies.
    INFO - step - Applying position restraints.
    INFO - step - Writing output.

AF-Q58295-F1-model_v3.pdb
47360
    INFO - general - Read 1 molecules from PDB file clean.pdb
    INFO - step - Guessing the bonds.
    INFO - general - 1 molecules after guessing bonds
    INFO - step - Repairing the graph.
    INFO - general - Applying modification N-ter to residue A-MET1
    INFO - general - Applying modification C-ter to residue A-LYS1634
    INFO - missing-atom - Missing atom ILE1454:CD
    INFO - step - Dealing with modifications.
 WARNING - unknown-input - Could not identify the modifications for residues [], involving atoms ['12063-CD1']
    INFO - general - Identified the modifications ['N-ter'] on residues ['MET1', 'MET1', 'MET1', 'MET1']
    INFO - general - Identified the modifications ['C-ter'] on residues ['LYS1634', 'LYS1634', 'LYS1634']
    INFO - step - Read input.
    INFO - step - Creating the graph at the target resolution.
    INFO - general - Applying modification mapping ('C-ter',)
    INFO - general - Applying modification mapping ('N-ter',)
    INFO - step - Averaging the coordinates.
    INFO - step - Applying the links.
    INFO - step - Placing the charge dummies.
    INFO - step - Applying position restraints.
    INFO - step - Writing output.

AF-B1GZ76-F1-model_v3.pdb
63178
    INFO - general - Read 1 molecules from PDB file clean.pdb
    INFO - step - Guessing the bonds.
    INFO - general - 1 molecules after guessing bonds
    INFO - step - Repairing the graph.
    INFO - general - Applying modification N-ter to residue A-MET1
    INFO - general - Applying modification C-ter to residue A-ASP1593
    INFO - missing-atom - Missing atom ILE198:CD
    INFO - step - Dealing with modifications.
 WARNING - unknown-input - Could not identify the modifications for residues [], involving atoms ['1614-CD1']
    INFO - general - Identified the modifications ['N-ter'] on residues ['MET1', 'MET1', 'MET1', 'MET1']
    INFO - general - Identified the modifications ['C-ter'] on residues ['ASP1593', 'ASP1593', 'ASP1593']
    INFO - step - Read input.
    INFO - step - Creating the graph at the target resolution.
    INFO - general - Applying modification mapping ('C-ter',)
    INFO - general - Applying modification mapping ('N-ter',)
    INFO - step - Averaging the coordinates.
    INFO - step - Applying the links.
    INFO - step - Placing the charge dummies.
    INFO - step - Applying position restraints.
    INFO - step - Writing output.

AF-A1ZA47-F1-model_v3.pdb
63383
    INFO - general - Read 1 molecules from PDB file clean.pdb
    INFO - step - Guessing the bonds.
    INFO - general - 1 molecules after guessing bonds
    INFO - step - Repairing the graph.
    INFO - general - Applying modification N-ter to residue A-MET1
    INFO - general - Applying modification C-ter to residue A-ARG2194
    INFO - missing-atom - Missing atom ILE609:CD
    INFO - step - Dealing with modifications.
 WARNING - unknown-input - Could not identify the modifications for residues [], involving atoms ['4568-CD1']
    INFO - general - Identified the modifications ['N-ter'] on residues ['MET1', 'MET1', 'MET1', 'MET1']
    INFO - general - Identified the modifications ['C-ter'] on residues ['ARG2194', 'ARG2194', 'ARG2194']
    INFO - step - Read input.
    INFO - step - Creating the graph at the target resolution.
    INFO - general - Applying modification mapping ('C-ter',)
    INFO - general - Applying modification mapping ('N-ter',)
    INFO - step - Averaging the coordinates.
    INFO - step - Applying the links.
    INFO - step - Placing the charge dummies.
    INFO - step - Applying position restraints.
    INFO - step - Writing output.

AF-J9VQ06-F1-model_v3.pdb
64705
    INFO - general - Read 1 molecules from PDB file clean.pdb
    INFO - step - Guessing the bonds.
    INFO - general - 1 molecules after guessing bonds
    INFO - step - Repairing the graph.
    INFO - general - Applying modification N-ter to residue A-MET1
    INFO - step - Dealing with modifications.
    INFO - general - Identified the modifications ['N-ter'] on residues ['MET1', 'MET1', 'MET1', 'MET1']
 WARNING - unknown-input - Could not identify the modifications for residues ['HIS439', 'TYR942'], involving atoms ['7475-OXT']
    INFO - step - Read input.
    INFO - step - Creating the graph at the target resolution.
    INFO - general - Applying modification mapping ('N-ter',)
    INFO - step - Averaging the coordinates.
    INFO - step - Applying the links.
    INFO - step - Placing the charge dummies.
    INFO - step - Applying position restraints.
    INFO - step - Writing output.

AF-F1QWK4-F1-model_v3.pdb
111262
    INFO - general - Read 1 molecules from PDB file clean.pdb
    INFO - step - Guessing the bonds.
    INFO - general - 1 molecules after guessing bonds
    INFO - step - Repairing the graph.
    INFO - general - Applying modification N-ter to residue A-MET1
    INFO - step - Dealing with modifications.
    INFO - general - Identified the modifications ['N-ter'] on residues ['MET1', 'MET1', 'MET1', 'MET1']
 WARNING - unknown-input - Could not identify the modifications for residues ['HIS982', 'ASP1673'], involving atoms ['13138-OXT']
    INFO - step - Read input.
    INFO - step - Creating the graph at the target resolution.
    INFO - general - Applying modification mapping ('N-ter',)
    INFO - step - Averaging the coordinates.
    INFO - step - Applying the links.
    INFO - step - Placing the charge dummies.
    INFO - step - Applying position restraints.
    INFO - step - Writing output.

AF-P64653-F1-model_v3.pdb
126549
    INFO - general - Read 1 molecules from PDB file clean.pdb
    INFO - step - Guessing the bonds.
    INFO - general - 1 molecules after guessing bonds
    INFO - step - Repairing the graph.
    INFO - general - Applying modification N-ter to residue A-MET1
    INFO - step - Dealing with modifications.
    INFO - general - Identified the modifications ['N-ter'] on residues ['MET1', 'MET1', 'MET1', 'MET1']
 WARNING - unknown-input - Could not identify the modifications for residues ['CYS43'], involving atoms ['339-OXT']
    INFO - step - Read input.
    INFO - step - Creating the graph at the target resolution.
    INFO - general - Applying modification mapping ('N-ter',)
    INFO - step - Averaging the coordinates.
    INFO - step - Applying the links.
    INFO - step - Placing the charge dummies.
    INFO - step - Applying position restraints.
    INFO - step - Writing output.

[pckroon]

Great! Very nice work.

There appears to be something fishy with C-terminal cysteines (AF-Q87KU6-F1-model_v3 and AF-Q5WN60-F1-model_v3). This is probably a data file issue on our side, but it could also be a strange bond involving one of the oxygens due to the conformation
AF-A3NBS1-F1-model_v3 and AF-P25063-F1-model_v3 is probably an iffy conformation where OXT is bridging between 2 residues.
The last three, AF-A3NA43-F1-model_v3, AF-A5EHJ0-F1-model_v3, and AF-Q05602-F1-model_v3 are probably a combination of conformation and data files: I think CD1 is too far away from CG to get a bond based on geometry; and the naming scheme is slightly different from what we use (e.g. CD1 vs CD).

Could you investigate the conformations of these at the problematic atoms? I'll take a look at our data files based on your findings. Consider writing out the bonds that MakeBonds finds using the -write-graph flag.

[Marco447]

I will have a look at it. I just realized that I provided the wrong proteins, it is now corrected in the first post.

[Marco447]

@pckroon I investigated it and the problem in all 7 cases is in the pdb files (missing or extra bonds, and switched atoms etc.)

[fgrunewald]

@pckroon I wonder, if we can issue a meaningful warning somewhere, if the atomistic bonded connections for protein AAs are too bad? For example, with the exception of Cysteine we know that the regular AAs shouldn't make connections over the side-chain. Cases where proteins structures are large and you have one bad contact are almost impossible to debug at the moment, unless you manually check thousands of bonds.

[pckroon]

We could do some basic (!) sanity checking, such as checking the degrees of atoms. Or makebonds should check for close contacts (since it has the distance matrix anyway). I don't think you want to be too thorough for the risk of it being very expensive. I do agree that we need to make the guru-meditation a bit easier.
Could you open an issue along these lines?

[pckroon]

Oh, and sanity checking the resulting CG model+conformation is also a great idea.