PDBFixer - in some cases added residues have amides with non-trans OCNH stereochemistry

[ivanicj]

Hi There,

I first want to say that for my cases PDBFixer is superior to other methods. For the case of 7zaw.pdb - Human GPC3 having missing atoms and residues (many) - PDBFixer is able to add the missing residues (that can be very long) extremely well, while other methods just aren't anywhere near as good. I noticed in some cases that linked residues can have amides with non-trans OCNH stereochemistry, and then MM minimization can actually make cis OCNH structures.

The test example I provide is 7zaw.pdb (RCSB) where PDBFixer produces the structure in 7zaw_PDBFixer. Examples of residue links having non-trans OCNH structures include:

3(G)-4(D)
5(A)-6(T)
21(L)-22(K)
28(P)-29(V)
161(D)-162(S)
162(S)-163(A)
284(R)-285(I)

I was wondering if there is a way to add residues whereby trans OCNH stereochemistry is enforced. I understand that the residues may have unfavorable positions but I'm hoping that a MM minimization will 'repair' them, while also keeping amide trans OCNH structures.

It's possible I'm not using PDBFixer correctly and thus I end up with non-trans OCNH amides, if so please let me know as this would be the easiest solution.

Thank you for any help you can offer, I very much appreciate it.

Joe

P.S. Note that I had to add ".txt" to end of pdb files so that github would upload them.

7zaw.pdb.txt
7zaw_PDBFixer.pdb.txt

[peastman]

I think I see the problem. To add missing residues, it first puts them in unrealistic positions, then energy minimizes with a soft-core force field to bring them into more realistic ones. The soft-core force field was created based on Amber ff99SB-ILDN. The problem is that the script used to create it didn't consider wildcards. It's missing any torsion term whose definition involves a wildcard. That includes the one to restrict rotation of the peptide bond.

That should be an easy fix.

[ivanicj]

If this is an easy fix I'll be very, very happy. Thank you very much again for looking at this!

[peastman]

The fix is in #308. Can you try it out and see if it works for you?

I tried to write a test case that would verify all added peptide bonds were in the right conformation, but it still found some cis ones, just not as many as before. I suppose that's expected: cis peptide bonds do happen, and anyway, we're just doing a local energy minimization, not a full equilibration. Do you have any suggestions on other ways to test it?

[ivanicj]

Great! Thanks! What is the best way to install version #308? Originally I installed OpenMM with conda and then installed PDBFixer also with conda - conda install -c conda-forge pdbfixer.

Should I download source #308, activate environment OpenMM, and then 'python setup.py install'? Will this replace the original PDBFixer? Sorry for the dumb questions, I just want to be sure to install #308 correctly.

[peastman]

Since this is prerelease code, it's best to test it in a new environment. The current PDBFixer code also requires recent changes to OpenMM, so you should install the 8.2 beta. You can do it like this:

mamba create --name testenv
conda activate testenv
mamba install -c conda-forge/label/openmm_rc -c conda-forge openmm

Next download the source for the PDBFixer branch containing the change, cd to the root directory, and type

pip install .

[ivanicj]

Thank you very much for the help with installing the test version. I think I did the right thing (git clone https://github.com/peastman/pdbfixer.git). I tested it on the 7zaw.pdb and the resultant pdb is attached (.txt added at end again so github could upload it). The missing residues were added at:

1 - 6
12 - 30
159 - 163
283 - 286
322 - 354
407 - 427
450 - 464

I noticed some cis/90 degree orientations at:

1(E) - 2(T)
4(D) - 5(A)

163(A) - 164(L) 90 degrees
283(Y) - 284(R) 90

322(A) - 323(H) 90
323(H) - 324(S)
324(S) - 325(Q) 90
326(Q) - 327(R) ~90
328(Q) - 329(Y)
330(R) - 331(F) 90
331(F) - 332(A) 90
336(E) - 337(D)
343(K) - 344(V)

415(N) - 416(Q)
418(N) - 419(L)
423(K) - 424(M)

458(K) - 459(H)
463(H) - 464(H)

I understand that this is a very tough case. As far as other ways to test it I can't think of any besides the current test.
Am I right in understanding that there is a local energy minimization of the added residues?

7zaw.pdb.txt
7zaw_PDBFixer.pdb.txt

[peastman]

It seems the torsion energies aren't enough to make the energy minimization fix them.

I tried implementing a different approach. When it first adds the residues, before minimizing them, it tries to rotate them to get the correct orientations for the peptide bonds. That helps more, though it still doesn't completely fix the problem. In 7ZAW, it brings the number of cis bonds down to only 7. I'll see if I can improve it any further then create a new PR for you to try out.

[peastman]

Try the version in #309.

[ivanicj]

I very much appreciate you working on this. I know that this is a very tough case (7zaw).
The latest results are in 7zaw_PDBFixer.pdb(.txt).

The missing residues were added at:

1 - 6
12 - 30
159 - 163
283 - 286
322 - 354
407 - 427
450 - 464

I took a look and there seems to be less cis HNCO amide links, two example of cis are:

20(G)-21(L)
21(L)-22(K)

I also notice some very distorted residue structures which is why I was wondering about the local energy minimizations.

Again, I very much appreciate you looking at this and recognize that this is a very, very difficult case.

7zaw_PDBFixer.pdb.txt

[peastman]

The energy minimization in PDBFixer uses a very approximate force field that just tries to put things into roughly the right positions. Before simulating, you should energy minimize with a real force field. Does that fix the distorted structures?

Overall would you say the results are an improvement from before, or is it better in some ways and worse in others?