[gmx-users] large error bars in PMF

[chris.neale at utoronto.ca]

I see now what you mean. As it happens, I doubt that this would have  
caused the problem since no force was applied on X and Y dimensions,  
so it would require that there was a PBC-based distance degeneracy  
along Z, although this is of course possible and hopefully Rebecca  
will answer this part.

Also, thanks for the pull_dimension/pull_vec fix.

Chris.

-- original message --


chris.neale at utoronto.ca wrote:

[Hide Quoted Text]
I don't see why the box-type makes any difference whatsoever. It is
possible that if you use a rhombic dodecahedron, you may reduce the
system size, thus simulate more ns/day, thus converge faster, but that
should be the only effect. I would be interested to hear more from
Justin about how the box-shape is expected to affect peptide rotation...
perhaps I misunderstand this point.
My point was not that the box shape has any effect on protein rotation.  That
will happen regardless of the box shape, of course.  My suggestion for  
this box
type was that since Rebeca has a system that is essentially spherically
symmetric (i.e. two proteins connected by some arbitrary vector, which are at
the same time rotating freely), then she must use a suitable box shape that
reflects this type of symmetry.  I never got a clear answer to whether or not
the weird interactions she cited were due to PBC or not, but if one uses a
rectangular box for a system like this one, there can be artificial  
interactions
very easily.

[Hide Quoted Text]
I have a few other comments.

1. If you allow the peptide to rotate freely, then you do indeed need to
converge all of their different rotational interactions. An alternative
is to apply orientational restraints during the pulling (assuming that
you know the bound state) and then to correct to an unrestrained state
at large separations. You can see, for instance,  D. L. Mobley, J. D.
Chodera, K. A. Dill. "On the use of orientational restraints and
symmetry number corrections in alchemical free energy calculations", ...

2. You are using "pull_dim = N N Y" which, if I haven't entirely
forgotten how the pull-code works, means that the distance along Z is
restrained but the distance along X and Y is free to change. What you
end up with by sampling in this way is pretty strange and will require a
really weird volumetric correction in the absence of infinite sampling
time. You must decide to either: (i) pull to a spherical distance:

pull_dim = Y Y Y
pull_geometry = distance

or (ii) to pull along a defined vector

pull_dim = Y Y Y
pull_geometry = direction
Just a note here - if you set direction geometry, the pull_dim keyword is not
used, but pull_vec is.

-Justin

[Hide Quoted Text]
What you have done:

pull_dim = N N Y
pull_geometry = distance

is only really useful when the system is isotropic along the XY plane
(at least in the time averaged sense), such as for a lipid bilayer, or
when the freedom in X and Y is very low, such as in a channel.

3. Finally, just because you sampled *more* doesn't mean that your
values are converged. Look into block averaging and test to see if your
binding free energy is drifting over time.

Good luck,
Chris.

-- original message --

Hi again,
I have one doub about the suggestion of using a dodecahedral box for my
umbrella sampling to remove the problems I am having with the peptides
rotating. I cannot see why a dodecaheral box is going to avoid this.
Would it be better a truncated octahedron?
Thanks a lot for your help.
Best wishes,
Rebeca.

<... snip...>