[gmx-users] REST variant of REMD

[Mark Abraham]

Anthony Armstrong wrote:
> Hello all,
> 
> I'd like to run REMD on a peptide system, but it turns out out my
> system is a little to large, and I've determined that to get a swap
> probablity of .2, a separation of temperature of only a couple degrees
> between replicas would be required.  I ran across the paper of Liu et al.
> (PNAS 2005 102:13749-13754) on Replica Exchange with Solute Tempering
> which addresses this problem and which appears to attempt temperature
> swaps based on a "modified" total potential in which solute/solute and
> solute/protein interactions are first scaled.

I'm also grappling with this kind of problem, and having looked at the 
above paper, and Affentranger et al JCTC 2:217 and Okur et al. JCTC 
2:420, I prefer Okur et al. That paper also criticises Liu et al for 
choosing a test system that they find to be almost insensitive to 
solvation treatment. They use an approximate Hamiltonian only during 
exchange, so while the simulations are no longer canonical ensembles, 
they are at least produced by normal MD Hamiltonians between exchange 
attempts.

> My question is:
> 
> How might I go about calculating a user defined energy (in this case
> E(X) = Ep(X) + A(T) * Ew(X) + B(T) * Epw -->  where X is the set of
> system coordinates, subscripts p, w, and pw denote protein/protein,
> water/water, and protein/water interactions, respectively, and A(T) and
> B(T) are scale factors involving the target temperature for the given
> replica) and then pass this energy rather than the unscaled potential
> to gromacs for determining whether or not to swap temperatures.

I'm currently debugging my implementation of the Okur method. You need 
to add a new potential energy term (unless you want to write over the 
old one) in idef.h, and rework the functions in repl_ex.c to calculate 
that new potential (if you can do the above decomposition), and also to 
do the exchange using that new potential energy.

Mark