Geometric optimization following amino acid substitution in PyMol

Question

[Hansen_869]

Hi!

I have a sequence of a scFv. I want to generate a representative 3D structure. If I try to find a structure using Alfafold og Swissmodel, they usually look pretty good, but some of the amino acids are off.

In Pymol I can substitute the wrong amino acids to the right ones, but the atoms tend to clash with the adjacent ones and the structure doesn't look right afterwards.

Is there any tool that can optimize the structure following amino acid substitution? Perhaps a software that can even do the amino acid substituion and optimize the geometry?

Thanks in advance!

Answer 1

[Mensur Dlakic]

Your way of making residue substitutions in PyMol and minimizing afterwards is suboptimal. If you still want to do it, AMBER will do the job.

https://ambermd.org/

Assuming there is a homologous protein of known structure, it should be much better if you leave the modeling to AlphaFold2. It will include the minimization at the end that will optimize both the backbone and side-chain positions. I don't see how you could do a better job by manually replacing residues + minimization than AlphaFold2 would do. Even SwissModel is likely to do a better job, despite your statement that "some of the amino acids are off." What does that even mean?

[Hansen_869]

Hi Mensur,

Thanks for your response! What I mean is that when I use Alphafold, I get structures that look good, but the amino acids are usually only at 90% identity with my inputtet sequence, so I have to manually replace them afterwards. Is there a way to generate a custom structure with Alphafold, with my exact amino acid sequence?

Thanks!

[Mensur Dlakic]

What I mean is that when I use Alphafold, I get structures that look good, but the amino acids are usually only at 90% identity with my inputtet sequence, so I have to manually replace them afterwards.

There is no such a thing as giving a sequence to AlphaFold2 and getting a model back that is 90% identical. Please consider this: what good would be a program that mutates residues at will? AF2 models exactly the sequence it is given, and so do all other 3D modeling programs.

Maybe you are using proteins modeled from the AlphaFold2 database and the best matches have only 90% identity to your sequence? If so, you will have to model the exact sequence. That can be done locally if you have the AF2 installation available, or you can try online:

https://colab.research.google.com/github/sokrypton/ColabFold/blob/main/AlphaFold2.ipynb

If you go with the latter option, I recommend that you select 5 in num_relax and pdb100 in template_mode dropdown menus.

[Hansen_869]

Hi Mensur,

I believe the ladder is what I've done. I have a sequence of a hypothetical scFv, which there does not yet exist a structure for. What I've done is going to https://alphafold.ebi.ac.uk/search/sequence/ and entered my sequence. All I get are proteins with sequences similar to my protein and not an actual custom model. From where can i download AF2 to generate the model locally? Can i generate models offline?

Thanks!

[Mensur Dlakic]

Installing AF2 is both time- and resource-expensive, and requires considerable expertise to run even if you manage to install the program. Based on what I perceive as your level of expertise, I would strongly advise you against that option. I suggest that you use the ColabFold service I listed in my previous post. But if you really want to give a try to local AF2 installation:

https://github.com/google-deepmind/alphafold

[Hansen_869]

Thank you! I used the ColabFold and it pretty much did the job. A follow-up question though: I have a heavy and a light chain. How would i best go about modelling them? If i model them in one continuous sequence, AF2 gives me some sort of fusion chain. If i model them separately and put them together in a software like Pymol, how would I know how they are supposed to be oriented to one another?