Adding parameters for non-canonical amino acids

[aminsagar]

Dear Sirah Developers/Users,
Is there some documentation on generating parameters for new residues like norleucine, for example?
I would be really grateful for any suggestions.
Best,
Amin.

[SIRAHFF]

Dear Amin,
thanks for reaching out. SIRAH parameterizations are based on interaction points that each residue or moiety establishes. Usually, parameters are derived from statistical information from the PDB or canonical structures. After a quick look at the PDB, it seems the side chain of NLE could be modeled with 2 beads placed at the CG and CE positions. We can create such residue for you. However, if you send us the PDB structure you are interested in, we could provide a more accurate guess. Additionally, are you using Gromacs or Amber for your research?
Best,
The SIRAH team

[aminsagar]

Thanks @SIRAHFF
I used NLE as an example but my broader aim is to parameterize hundreds of non-canonical amino acids in SIRAH. Many of these might not exist is PDB. I am wondering if there is a way to generate approximate SIRAH parameters using all atom MD simulations using an approach similar to https://github.com/GMPavanLab/Swarm-CG

Best,
Amin.

[SIRAHFF]

Hi Amin,
I couldn't open the link, but I think I know the work. I misunderstood your question. SIRAH uses a top-down approach, so, we never explored the generation of parameters from all atom MD.
Another point of concern could be the naming scheme for the new beads/residues, which is currently quite limited in AMBER. So, the compatibility with exisiting parameters could be an issue.
Anyhow, we would be happy to discuss that if you think it might help.
Best,
Sergio

[aminsagar]

Thanks @SIRAHFF

Sorry, I updated the link in my previous comment.

I think an approach to develop approximate SIRAH parameters from All Atom simulations for peptoid/peptide research could be quite useful especially because the higher resolution of the backbone in SIRAH. Coarse grained peptide folding simulations, for example, are not possible with Martini because of the need for secondary structure restraints, which can't be applied for flexible peptides with transient secondary structures.

Can we try to do an experiment where we take a non-canonical amino acid which is present in PDB, like norleucine, and try to develop SIRAH parameters using both the top-down approach and an all-atom MD based approach to see if we arrive at somewhat similar parameters?

Best,
Amin

[SIRAHFF]

Hi Amin,
yes sure, it could be an interesting test. We should define the molecular system, MD engine and simulation conditions for the test.
Notice, we have had some lucky cases with peptide folding, but I don't think SIRAH will perform well for folding Ad-initio. Still a comparative set of simulations would be something interesting to check.
What you have in mind?
Sergio

[AledHe]

Hello Amin and @SIRAHFF,

I am also interested in this topic, although I am not a researcher in MD or in the use of AMBER or GROMACS, so I lack a basic understanding of the concepts of MD and quantum chemistry. but I hope I can express my viewpoint clearly.

As far as we know, more than 600 SMILES structures have been recorded for various non-canonical amino acids (as of 2018, according to the RESID Database), such as the previously mentioned norleucine. Apart from post-translational modification (PTM) amino acids, which somewhat resemble their "parent" structures, some very exotic amino acids exist. For example, N-(2,2-diphenylethyl)-glycine {C1=CC=C(C=C1)C(CNCC(=O)O)C2=CC=CC=C2} is difficult to assess in terms of whether it still retains similar properties to the original amino acids.

As described in this paper, a common practice is to build a pseudo-dipeptide, ACT-X-NME, and then optimize its geometry. However, this involves a complex series of procedures that can be quite perplexing and seem computing intensive for me, not to mention the challenge of simulating a 10-50 amino acid peptide (using AMBER or GROMACS, I assume?). I also don’t fully understand if all amino acids used need to have their parameters defined in the same force field to build the peptide.

As Amin suggested, proposing an end-to-end solution could have a significant impact on the field, or a potential solution. this is also a problem that deep learning models cannot yet solve, despite the fact that AlphaFold3 supports 23 PTMs and appears to be expanding its capabilities (though they have stopped releasing the code).

[SIRAHFF]

Hi again,
yes, totally agree with that. Indeed, these guys have shown and other people confirmed that SIRAH simulations are quite accurate for DDBinding energies, so it could be possible to estimate in a relatively fast way the effect of the non-natural aminoacids.
Unfortunately, we are short of hands to address this at the moment, but will be certainly happy to help if anyone is interested.
Best,
Sergio

[aminsagar]

Hello. Sorry for the delay.

One of my objectives is also to use SIRAH for estimating the changes in binding affinities associated with mutating natural amino acids to non-canonical ones.
Another one is folding simulations of short (10-20 aa) peptides.

I am not aware of a binding affinity dataset which has non-canonical amino acids. If you are aware of such datasets, it would be great to benchmark SIRAH vs all-atoms simulations on it.

For folding simulations, there are some general trends that we can try to reproduce. For example, Aib is known to increase helicity. 1-aminocyclohexane-1-carboxylic acid can increase γ-turn conformations in small peptides. 1-aminocyclopropane-1-carboxylic acid prefers bridge.

Many such modifications are discussed in https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10296201/

I can propose some concrete systems next week. I think relaxin could be a good system to test. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7602263/

I would like to use gromacs for the simulations, either classical or replica exchange.
Best,
Amin