Re: PMF from ABF simulation not plateauing

From: Jérôme Hénin (jerome.henin_at_ibpc.fr)
Date: Thu May 08 2014 - 17:37:20 CDT

Hi Riley,

You don't give details on what exactly happens to your PMF and your simulation, but if I were to try a guess, I'd say that the PMF doesn't converge because the dimer gets stuck in dimer form, and you don't sample larger distance, or not well.

What is happening is that the restraints affect the actual PMF, but due to implementation details they don't affect the computed PMF (that's because the external restraint forces are not counted as part of the system forces - it's a matter of deciding what the relevant potential energy function is).

The cylindrical restraints break the spherical symmetry of the distance coordinate, the 3d coordinate becomes effectively 1d. The net result is a bias towards shorter distances. They are not strictly orthogonal to the distance coordinate.

The best approach would be to replace the distance coordinate with a distanceZ coordinate that matches the actual system geometry. Then the external bias will actually be orthogonal to the ABF coordinate.

Best,
Jerome

----- Mail original -----
> Hello all,
>
>
> I am trying to generate a PMF using ABF and am having trouble getting the PMF
> to plateau. I have hideJacobian set on.
>
>
> I use the selectConstraints x and y commands to restrict two water molecules
> to the z-coordinate. I am trying to generate a dimerization of the two water
> molecules from a specific conformation, thus the constraints. I am
> performing the ABF on a distance colvar between the oxygens of the two
> waters.
>
>
> When I run such a simulation without the constraints, I get a lovely smooth
> water-water PMF that plateaus at around 3.5 kcal/mol. Keep in mind I am
> setting hideJacobian to on, so the PMF should plateau. I have read the NAMD
> information on ABF and am aware that constraints on atoms involved in the
> distance coordinate can disrupt the accuracy of the PMF. My thought was that
> since the x and y constraints are orthogonal to the distance colvar, that
> might not happen.
>
>
> Can anyone with expertise in ABF lend any insight on why this happens and how
> I might circumvent the problem? I realize that I could use umbrella sampling
> or some other method of measuring the PMF, but I have performed some other
> related calculations with ABF and would like to use it with this one as
> well, if possible.
>
>
> Thanks for any and all suggestions.
>
>
> Riley Workman
>

This archive was generated by hypermail 2.1.6 : Wed Dec 31 2014 - 23:22:23 CST