[gmx-users] FEP in vacuum simulation - inconsistent results among FEP methods

[Melania T]

Hello Melania,

The different methods for calculating Free Energy should be reasonably
close to one another. Since you are mutating one molecule into another I am
betting that some of the bonds are mutating as well and this isn't a simple
"turn off the intermolecular interactions" scenario. If this is the case it
is important that you do not constrain the mutating bonds with LINCSm. I
suggest using h-bonds rather than all-bonds in your LINCS constraints.
These kind of strange results occur when you are trying to constrain a bond
as it is mutating from one equilibrium length to another. You may need to
reduce your timestep, but chances are 2 fs will be alright if you are
constraining the hydrogen bonds.

I recommend reading "An Introduction to Best Practices in Free Energy
Calculations" by Mobley and Shirts. They discuss this and much more as well.

Good luck,

Your future self,

Melania

On Tue, May 28, 2019 at 7:29 PM Melania T <melaniacompchem at gmail.com> wrote:

> I am calculating the free energy change in transforming one small molecule
> into another by the simple removal of a CO2 from the first molecule, and
> also mutating some bond parameters slighty since the second molecule has a
> couple different atom types in GAFF without the CO2 attached to it.
>
> I analyze the results using alchemical_analysis.py
>
> In solution, the change in free energy is consistent among all 6 default
> methods used in alchemical_analysis.py (TI, TI-cubic, DEXP, IEXP, BAR,
> MBAR).
>
> However in vacuum after 100 ns, the results are still very different among
> the methods. I use 10 coulomb windows and 20 LJ/Bond windows for both
> simulations (evenly spaced). According to the dF_t plot printed by
> alchemical_analysis.py the results are converged for MBAR for pretty much
> the entire simulation. Gromacs/2019.1
>
> (units in kJ/mol)
>
> TI : 0.284 +/- 0.014
> TI-CUBIC: 1.483 +/- 0.015
> DEXP: -7.196 +/- 0.016
> IEXP: -7.290 +/- 0.014
> BAR: -7.236 +/- 0.009
> MBAR: -6.110 +/- 0.008
>
> Is this common in vacuum simulations or am I likely doing something wrong?
>
> is it normal to have to use more lambda windows in a vacuum simulation
> than in solution?
>
> Thanks,
>
> Melania
>