Questions about new flags

[qsabanes]

Hey, thanks so much for this tool. I can manage to run smoothly rbfe calculations with ligands that was unfeasable with just ATM.

I do have some questions about the new necessary flags, I tried to understand them based on the inputs you provide but I want to be sure if I'm correct.

Next examples are related to the use of ATS for peptides and similar ligands. I have not tried with small molecules yet.
LIG_OFFSET : I do not know the meaning of this one, I have seen it in other examples of ABFE and in the code but I don't really get what it does. I have it off in my systems.
RCPT_CM_ATOMS : Only receptor atoms around the modified area are selected now right?
LIGAND1_CM_ATOMS and LIGAND1_ATTACH_ATOM: I believe is the CA of the modified residue?
ALIGN_LIGAND1_REF_ATOMS: Only selecting atoms in the modified res?

I also wanted to ask. ATM should completely be unfeasible with peptides right? I have tried to run the same system with ATM (to compare differences) but I always got several crashes and/or a huge predicted error during uwham.
Also, have you seen a difference in speed performance? Is ATS slower than ATM or it should have the same speed?

Many thanks!

[egallicc]

Sorry Quico. I missed your questions until now:

LIGOFFSET (bad historical name) is the displacement applied to the binding site restraint center. It is added to the centroid of RCPT_CM_ATOMS before its distance from the centroid of LIG_CM_ATOMS is calculated (for ABFE). In RBFE calculations, it is added to the centroid of RCPT_CM_ATOMS for LIG1_CM_ATOMS, and to RCPT_CM_ATOMS+DISPLACEMENT for LIG2_CM_ATOMS. It is used to locate the binding site region more easily and precisely when the centroid of RCPT_CM_ATOMS is unsuitable (such as for a shallow binding site). It is also used in ABFE calculations when starting from the ligand in solution.

RCPT_CM_ATOMS : Only receptor atoms around the modified area are selected now right?

How to best define the binding site region in the case of protein-peptide and protein-protein complexes is an open question in my mind. The Boresch-style external degrees of freedom of the ligand relative to the receptor are rigorous, but I have doubts about their ability to capture the physical/biological significance of the "bound" state of the protein-protein complex. In the calculations here, I opted to turn off the binding site restraints and impose flat-bottom positional restraints on some of the C-alpha atoms of the peptide. So, I really don't know what the best way is to set RCPT_CM_ATOMS, etc., in ways that are useful for peptide/protein design.

LIGAND1_CM_ATOMS and LIGAND1_ATTACH_ATOM: I believe is the CA of the modified residue?

Yes.

ALIGN_LIGAND1_REF_ATOMS: Only selecting atoms in the modified res?

Yes, I picked the C-alpha, N, and C backbone atoms of the mutated residues.

I have tried to run the same system with ATM (to compare differences) but I always got several crashes and/or a huge predicted error during uwham.

Right. I think it is hopeless beyond 15/20 atoms.

I have not tracked it, but the time-step speed of ATS should be the same as that of standard ATM. There should be a way to save time by using fewer lambda states for congeneric ligand RBFEs,

[qsabanes]

Hi Emilio,

thanks for the answers ans sorry for the delay. I could come back to this recently.

I had one more question. Have you considered how to tackle with ATS if the ligand is perturbed in two different regions of the protein? All flags would be the same but LIGAND_ATTACH_ATOM is where we are doubting how to consider. Or just leave it and do the perturbation in two steps?

[egallicc]

Have you considered how to tackle with ATS if the ligand is perturbed in two different regions of the protein? All flags would be the same but LIGAND_ATTACH_ATOM is where we are doubting how to consider.

The ATMForce API allows each atom's displacement to be set independently. Thus, each atom of the variable region could have its own "attachment" atom. AToM-OpenMM's implementation supports only one variable region and one attachment atom, but it could easily read in multiple variable regions and attachment atoms. We should probably wait until the new ATMForce API is finalized before implementing this option.

Or just leave it and do the perturbation in two steps?

Yes, applying mutations in sequence could be more efficient. Coupled mutations (such as when a charge-changing mutation of one residue is offset by an opposite one nearby) could be an exception.