[gmx-users] Free Energy calculations of peptide-protein binding

[chris.neale at utoronto.ca]

> I am investigating a certain peptide which is well bound to some
> protein; their configuration is known from the PDB.
>
> My aim is to compare the binding free energies of several variants of
> the peptide.
>
> In particular, I wish to mutate some amino acids to others.
>
> I have read some tutorials and mailing list messages; still, I have
> several questions unanswered. I plan to use the Thermodynamic  
> Integration (TI) method in an
> "alchemical" setting, in which I "turn off" the side chain.
>
> The "naked" amino acid will then be my reference point comparing to real
> amino acids.
>
> In this way I should be able to obtain, using the usual thermodynamical
> cycle, the binding free energy difference of two amino acids (Delta
> Delta G), which is sufficient for my needs and should be more accurate
> than computing absolute binding free energies (Delta G).

correct.

>
>
>
> 1) Does this sound like the best approach?

That part's up to you

>
> 2) In particular, is it O.K. to use a "naked" amino acid (with no side
> chain at all) as the reference, or should Glycine or Alanine be used?

You want to have as little change as possible. I can't be scpecific  
here because you're trying to hide your system of study -- perhaps  
unavoidable in your case, but in general this is a surefire way to get  
poor advice.

e.g. Adding an entire TRP SC at once (or worse, ARG) is going to give  
you convergence problems -- be sure to get a good measure of  
convergence.

>
> 3) Following the above procedure, I still would not achieve a "naked"
> amino acid.
>
>     Rather, a side chain "ghost" remains, non-interacting but still
> bonded to itself and to the C-alpha.
>     Is it correct to assume that the contribution of such a "ghost"
> would cancel between the free and bound peptide?

Don't think about cancellation (since that would of course involve  
sampling). Better to recall that the free energy difference to not  
change the state is always zero. Therefore if you have identical  
(interacting) states, then you have no free energy difference.

>
> 4) Are the OPLS-AA force field, theTIP-3P water model and the NpT
> ensemble good choices?

Again, FF is up to you and hard to comment on without knowing what you  
are doing. If you use the Npt Ensemble, you should really also  
calculate any pressure work. If you use the NVT ensemble, you avoid  
this, but then your densities might not be perfect. Probably many  
people ignore this part, but its good to at least keep it in mind.

>
>
>
> I have seen two methods used to make the change between the initial and
> final states.
>
> The first (and simplest) is to use the couple-moltype parameter of
> mdrun.
>
> However, this seems to change a whole molecule, while I'm interested in
> changing only a part (the side chain).
>
>
>
> 5) Can I define the side chain and the amino acid backbone as different
> molecules, and change the former, still connecting them one to the
> other?

I dunno, sounds fishy.

>
>
>
> Alternatively, the topology of the B state can be explicitly provided
> (as described e.g. in section 5.7.4 of the version 4.0 user manual).
>

This is the way to go. But, again, be careful switching off 20 atoms  
at once, and be careful of charges.

>
>
> 6) The atom charges can be specified for state B, as appropriate for
> turning off the Coulomb interactions.
>
>     In order to turn off the vdW ones, must I define new atom types?
>
> 7) Are there perhaps such amino acid variants already built?  
> Finally, some technicalities:
>
>
>
> 8) I have seen the sc-alpha parameter (when sc-power = 1) given the
> values 1.0, 0.7 and 0.5. What is recommended?

This one has been discussed regularly on the gmx mailing list. You can  
find the answer there.

>
> 9) Should I employ DispCorr = EnerPres ?
>

Absolutely. Although keep in mind that this won't solve everything --  
we're still waiting on a particle Mesh LJ scheme for infinite LJs.

>
>
> Thanks,

Chris.