From: Jerome Henin (jhenin_at_cmm.chem.upenn.edu)
Date: Mon Dec 11 2006 - 11:29:59 CST
On Monday 11 December 2006 02:20, satya work wrote:
> Hello all,
> I am interested in using FEP, as implemented in NAMD, to compute free
> energy difference (1) between two ligands (ethanol->butanol) that bind to a
> protein (2) mutating a particular residue in a protein. I intend to perform
> (1) by mutating ethanol->butanol in water & in the presence of protein to
> complete the thermodynamic cycle. for (2), I am trying to understand the
> example presented in the tutorial (may be more questions will follow)
> But I am slightly confused about some fundamental aspects of implementation
> of FEP:
> 1. what are the interactions present in the system in the case of
Let us call the initial group A, the final group B and the rest O.
- all bonded terms: unmodified
- nonbonded O with O: unmodified
- nonbonded A with O: scaled by (1-lambda)
- nonbonded B with O: scaled by lambda
- nonbonded A with B: excluded
> 2. are the intra-bond interactions of individual ligands still present? i
> mean are the ligands still 'bonded together' at end points?
I am not sure what you mean, but I hope point 1. answers this.
> 3. are only the non-bonded interactions (LJ & Coulomb) scaled with
> 4. can FEP as implemented in NAMD be used in a case where there are no
> 'common chemical groups' between two ligands?
Yes, it can. It introduces a problem, though, because at each end-point, one
of the ligands is totally decoupled from its environment, and tends to move
around uncontrollably, making convergence of the FEP average a challenge. The
most rigorous way to deal with it that I know of consists in applying
positional restraints to keep the ligand roughly in the right spot, and
computing the free energy contribution of these restraints separately.
In cases where there *is* a common, unperturbed group, the bonded interactions
have that nice effect of keeping everything in place.
> 5. for a case of mutating residue X -> Y in a protein embedded in lipid
> environment, what thermodynamic cycle is best suited (is it even advisable
> to address this problem in case of a membrane protein, with all additional
> complexity and probably long time scales?)
It depends of your computational resources and on the "glassiness" of your
system. It it is very "soft" and will slowly relax here and there, the cost
for converging the calculation will be prohibitive. If it is fairly robust,
and there is not that much conformational space to be sampled, then you have
a good chance. It worked for us, among others (Henin et al, 2006, Biophysical
You might also want to consider a simplified or truncated model, if it is
enough to account for the environment around the ligand.
> 6. if the mutation is from a neutral to charged amino acid, what is the
> best way to create the hybrid topology with respect to charges? ( iam
> assuming i cannot use the charges 'as is' from CHARMM parameter set)
You can have a look at the VMD plugin called Mutator, which now does that kind
of dual-topology setup automatically. I would recommend doing it by hand with
PFSgen, though, but you can use the library of dual topologies that is
distributed with it:
Note that these do not include CMAP terms yet, so if you want to use the
latest and greatest of CHARMM for proteins, you can use our topology as a
template and just add CMAP terms.
Also, the change in total charge as the mutation occurs is an issue. Our
approach to that was to neutralize the charge by a counterion that was
mutated at the same time as the ligand. That counterion was restrained to the
bulk aqueous solution, so that the solvation term for it cancels out with the
equivalent transformation in water.
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:44:15 CST