Re: [AMBER] High standard deviation in entropy and decomping calculation

From: Jason Swails <jason.swails.gmail.com>
Date: Tue, 29 May 2012 10:50:38 -0700

I've already addressed the normal mode standard deviations in my response
to the other thread (which you contributed to as well). The patch for that
is attached to that response.

Bill's email is correct about these issues being addressed in AmberTools
12, so I will simply explain the large standard deviations for
decomposition analysis in AmberTools 1.5 here. The large standard
deviations (for the DELTA decomposed values only) result from how the std.
devs are calculated.

Regardless of whether a single or multiple trajectory protocol is used, the
standard deviation is calculated by the square root of the sum of squares
of the standard deviations of the complex, receptor, and ligand (that is,
it's propagated as though the std. devs. of the different species are
completely uncorrelated). This was done because it was determined that
storing entire arrays of all terms for all snapshots could require too much
memory (indeed, in one of our tests for pairwise decomposition we consumed
> 8 GB of RAM). Therefore, we just keep a running sum and running sum
squared for each term (so the test that used to take > 8 GB of RAM now
takes ~512 MB of RAM instead). The way we parsed the file made another
approach difficult to adopt.

For single trajectories, though, snapshots are, more or less, perfectly
correlated between the complex, receptor, and ligand (since the
conformations are the same in each state). Therefore, the variance
calculated as a sum-of-squares will drastically overestimate the variance
calculated from a population of DELTAs (that is, calculating the standard
deviation of the DELTA values for each snapshot).

In AmberTools 12, the decomposition parser was completely rewritten so that
we could keep a running sum and sum-of-squares of the DELTA values as well.
 This way, the memory requirements were kept at a minimum and we could
provide more realistic values of the standard deviation for the DELTA
section of the decomposition analysis.

HTH,
Jason

On Tue, May 29, 2012 at 5:53 AM, Cao Yang <muxiachuixue.163.com> wrote:

> Hi everybody,
> I came across a problem in entropy and decomping calculation with MMPBSA.
> The standard deviation is abnormally high in both case.
>
> 1. High standard deviation in entropy calculation
> The entropy of 20 frames were calculated by NMode with the following input
> file:
> Input file for running entropy calculations using NMode
> &general
> startframe=201, keep_files=2,
> strip_mask=":WAT:Cl-:CIO:Cs+:IB:K+:Li+:Na+:Rb+"
> /
> &nmode
> nmstartframe=1,
> nminterval=10, nmode_igb=1, nmode_istrng=0.1,
> /
> The result is as following:
> Complex:
> Entropy Term Average Std. Dev.
> -----------------------------------------------------------
> Translational: 16.7223 0.0000
> Rotational: 17.0094 0.0028
> Vibrational: 2452.2343 4.1753
> Total: 2485.9660 4.1757
>
> Receptor:
> Entropy Term Average Std. Dev.
> -----------------------------------------------------------
> Translational: 16.7119 0.0000
> Rotational: 16.9984 0.0026
> Vibrational: 2432.7994 3.1862
> Total: 2466.5098 3.1868
>
> Ligand:
> Entropy Term Average Std. Dev.
> -----------------------------------------------------------
> Translational: 12.7647 0.0000
> Rotational: 10.2737 0.0000
> Vibrational: 15.7722 0.0002
> Total: 38.8106 0.0002
>
> DELTA S total= -19.3544 +/- 86.2259
>
> -------------------------------------------------------------------------------
> I use AmberTools 1.5 with all bugfix. The total standard deviation is
> extremely high. Maybe AmberTools 12 could get over this? Any suggestions?
>
> 2. High standard deviation in decomping calculation
> Input:
> Per-residue PB decomposition
> &general
> startframe=201, endframe=500, verbose=1,
> /
> &pb
> istrng=0.100, cavity_surften=0.03252, cavity_offset=0.92,
> /
> &decomp
> idecomp=1,
> dec_verbose=1,
> /
> Output:
> Total Energy Decomposition:
> Residue | Location | Internal | van der Waals |
> Electrostatic | Polar Solvation | Non-Polar Solv. | TOTAL
> VAL 1 | R VAL 1 | 0.000 +/- 4.630 | -0.000 +/- 1.350 | -0.095
> +/- 20.783 | 0.095 +/- 19.466 | 0.000 +/- 0.181 | -0.000 +/- 28.882
> ASP 379 | R ASP 379 | 0.000 +/- 2.416 | 1.127 +/- 1.622 | -12.206
> +/- 11.846 | 6.517 +/- 8.712 | -0.082 +/- 0.020 | -4.644 +/- 14.990
> I use AmberTools 1.5 with all bugfix. The total standard deviation is
> extremely high. Any suggestions?
> Many thanks!
> Cao Yang,
> Zhejiang University
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>



-- 
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
352-392-4032
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue May 29 2012 - 11:00:03 PDT
Custom Search