Hi,
I started playing around with the gist functionality in cpptraj and thought
it would be a good idea to try to reproduce the original cucurbituril
results of Nguyen, Young and Gilson from JCP 137, 044101 (2012).
In general, this works nicely, after 500ns of standard simulation, with in
total 5000 snapshots, I get plots that look very similar to what Nguyen et
al report in their paper.
However, when looking at the Eww-norm.dx data, I see a number of voxels that
have a extremely high energy. As a first attempt, I have used a 21x21x21
grid with a spacing of 0.5 (this doesnt even encompass all of the
cucurbituril) - a histogrm of the Eww data shows a somewhat skewed, but
still approximate Gaussian distribution for values below 0, a large peak for
Eww values equal 0, but then 8 voxels with energies up to 300 kcal/mol. The
neigboring voxels of these high-energy ones have what I would consider
"normal" values.
This behavior becomes extremely pronounced when I use a larger grid for my
gist calculation (for a 41x41x41 grid with a spacing of 0.5A, I get energy
values up to 1E19, and about 13000 of the 68921 data points have a Eww-norm
> 100 kcal/mol.
My simulation box is about 35x35x35A, i.e. for larger gist grids one comes
close(r) to the edge of the box, so there might be some edge/PBC effects
(really??), especially as these high energy voxels tend to concentrate at
the edges of the gist box, but the high energy voxels of the 21-grid also
are high energy in the 41-grid, i.e. in the middle of the grid and far from
the edges of the larger grid.
I have played around with my trajin settings a bit, reading in only parts of
the trajectory, or using only every Nth snapshot, and although I do see some
differences, the same problematic voxels tend to show up again and again.
As I originally suspected that it could be statistical noise when voxels
arent populated often enough, I repeated the simulation and wrote out 250000
snapshots - this even increased the problem. I also tried NVT vs NpT, to no
avail.
Now my questions:
1) has anybody else seen this behavior (I have seen one post last year, but
there were no answers) or could there be something wrong with my simulation
or my analysis (well, this could alwas be the case i suppose) ?
2) is this normal and expected behavior?
2) if so, why this happens? Just some weird configuration of one/some of the
snapshots? Numerical instability? A problem of the gist implementation in
cpptraj? other?
3) what to do about this? First of all these voxels clutter my
visualization, but it is also difficult if I want to do some quantitative
analysis with these grids...can I always expect Eww to be below 0 (I suspect
no)? Should I try to smooth the grid, e.g. by replacing the high energy
values by the average of the surrounding 26 voxels?
Any thoughts or help appreciated!
Th.
PS. Ive run Amber14 pmemd.CUDA, cpptraj version is 15.00, the input file is
parm cb7_solv.top
trajin cb7_solv.crd
trajin prod.traj.nc 1 5000
autoimage origin
rmsd CUC :1 first
gist gridcntr 0. 0. 0. griddim 21 21 21 gridspacn 0.5 out gist_21_05.out
run
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Mon Aug 10 2015 - 03:00:02 PDT
