Hi Helmut and Dave,
My group has developed and been testing lambda-dependent restraints that
Tai-Sung Lee has implemented in pmemd.cuda, and if we get it fully
validated, it will be integrated into the next Amber24 release. These
lambda-dependent restraints have been used to implement more efficient
Boresch-type restraints used to restrain ligands in their
non-interacting dummy state to remain bound to protein targets in
absolute binding free energy (ABFE) calculations. Paper is forthcoming
- Abir Ganguly may talk about this next week at the Amber meeting.
Note: In an ABFE simulation, one transforms a real ligand state that is
interacting with the protein and solvent into a non-interacting "dummy"
state. Unrestrained, such a non-interacting dummy state of the ligand
would drift out of the binding pocket and disrupt replica exchange (and
ACES). Hence, on can use a special set of restraints, designated
"Boresch-type restraints" between the non-interacting ligand and the
protein that has the special property that they do not introduce a net
potential of mean force on any of the protein atoms (and thus will
produce the same protein ensemble as is the restraints were not present
at all). An analytic correction can be made for the free energy of
imposing the Boresch-type restraints on the non-interacting dummy state
of the ligand (to remove the free energy contribution of restraining
it), but this is not rigorous to apply to the interacting (real) state
of the ligand. Hence, if one uses a fixed (NOT lambda-dependent) set of
restraints throughout an alchemical transformation, one would have to do
a separate explicit sets of simulations to obtain the free energy
contribution of releasing ("loosening") the restraints from the
interacting ligand.
An alternative approach is to not have any restraints on the interacting
ligand at all (as it will stay bound to the protein due to these
interactions), and only introduce the lambda-dependent Boresch-type
restraints as the ligand transforms into the non-interacting dummy
state. In this way one avoids the need for a separate sets of
simulations altogether.
The implementation of lambda-dependent restraints is one of several very
useful new features for alchemical free energy simulations that will be
part of Amber24.
As an alternative, one can conduct a series of simulations using
different restraint force constants (with or without REMD) and use
parameter interpolation thermodynamic integration in a post-processing
step for the analysis - see:
A GPU-Accelerated Parameter Interpolation Thermodynamic Integration Free
Energy Method
Timothy J. Giese, Darrin M. York
J. Chem. Theory Comput. (2018) 14, 1564–1582
DOI: 10.1021/acs.jctc.7b01175
Dave's caution about potential caveats are certainly on point. I hope
this is helpful.
best wishes,
Darrin York
On 2/23/24 14:15, David A Case via AMBER wrote:
> On Thu, Feb 22, 2024, Helmut Carter via AMBER wrote:
>
>> Subject: Thermodynamic Integration of Loosening Restraints
>>
>>
>> I would like to determine the free energy change associated with
>> loosening
>> restraints on a small molecule, going from 100 kcal/mol*A^2 to 0.
>> Because
>> this alchemical transformation does not involve a change in the molecule
>> topology, I do not see a straightforward way of doing it in AMBER/pmemd.
>
> Some coding would be needed here. Remember that TI basically computes
> the
> average of dV/dlambda and various intermediate lambda values. Restraints
> are computed in ene.F90 (subroutine xconst) in sander. You would need to
> modify this code to make the restraints lambda-dependent, then compute
> the
> average value over a simulation. For your case, you don't need to
> specify
> icfe, since there is not need to consider two different prmtop files.
> You
> would just run a straightforward simulation and print out (or average
> on the
> fly) the required value.
>
> (Same idea for pmemd, but I'd try in sander first. If you want GPU
> acceleration, you would need to code a GPU kernel for that. So, consider
> exploring the idea on the CPU version first.)
>
> A caveat: I've never done this, and I might well be missing something.
> Certainly, you would need to be careful if there are large
> conformational changes upon relaxing the restraints -- e.g. if the small
> molecule dissocates from the protein when the restraints go to zero.
>
> Also: are you sure that what you describe is what you really want to
> do? The
> "usual" procedure is to decouple the small molecule from the rest of the
> system while restraints are still in place. After that, you can
> compute the
> free energy of releasing the restraints analytically, since the small
> molecule no longer interacts with the rest of the system. But your
> approach
> might have some advantages; I'm just suggesting that you think carefully
> about what you will do with your result once you obtain it.
>
> ...good luck....dac
>
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Feb 23 2024 - 13:30:02 PST
