Re: eABF 3D sampling using distance Colvars + X and Y: averaged PMF difference to 1D sampling

From: Jérôme Hénin (jerome.henin_at_ibpc.fr)
Date: Fri Aug 27 2021 - 07:05:16 CDT

Hi René,

----- On Aug 26, 2021, at 8:52 PM, René Hafner TUK <hamburge_at_physik.uni-kl.de> wrote:

> Hi Jérôme!
>> The way I extract it was to run second abf instance with only the
>> distanceFromMembrane colvar that is **not** applied.

>> Which turned out to be the same as when just weighting the zgrad(x,y,z) by the
>> respective counts and average over X,Y.

>> But we were wondering whether this is the correct way.
I don't think that is correct, because the simulation is biased in the directions X and Y, so you are not computing an unbiased average at each z value.

>> I even tried an easier setup with an almost flat membrane surface where I can
>> compare results to a simple distanceZ variable.

>> Here the distanceMin 1D case returns 4.6 (+. 0.1) kcal/mol (same as distanceZ)

>> the 3D case returns dG = 4.3 (+- 0.1) kcal/mol (with the above assumption)

>> Though what you suggest is to calculate the marginal distribution via the
>> following?

Just an integral instead of an average:

>> What I additionally needed to do in order to gain a smooth curve here was to
>> weight every grid point by its counts (such that "bad" guesses with low counts
>> get suppressed). But I am not yet the sure the scale is correct.
With the expression above you won't need to worry about that.

Jérôme

This archive was generated by hypermail 2.1.6 : Fri Dec 31 2021 - 23:17:11 CST