# AMBER Archive (2005)Subject: Re: AMBER: big fluctuation in MM-PBSA/GA

From: Eric Hu (eric.y.hu_at_gmail.com)
Date: Thu Apr 14 2005 - 19:21:40 CDT

Here we are talking about a small difference between two big numbers
which fluctuate so much that the difference is not significant any
more at each individual snapshot. I agree that the standard error in
the mean is probably more useful from the perspective of statistics
and probability. In this case the SD for PBTOT is 0.22 kcal/mol.
Therefore the PBTOT can be reported as -9.4 +/- 0.2 kcal/mol. What
about the confidence level? I assume that the distribution curve is
really flat...

Eric

On 4/14/05, David A. Case <case_at_scripps.edu> wrote:
> On Thu, Apr 14, 2005, Eric Hu wrote:
> > # DELTA
> > # -----------------------
> > # MEAN STD
> > # =======================
> > ELE -14.20 28.32
> > VDW -37.03 7.62
> > INT 0.01 0.02
> > GAS -51.21 32.31
> > PBSUR -4.00 0.76
> > PBCAL 45.81 32.31
> > PBSOL 41.82 31.85
> > PBELE 31.61 9.06
> > PBTOT -9.40 6.93
> > GBSUR -5.75 1.10
> > GB 41.98 27.46
> > GBSOL 36.23 26.85
> > GBELE 27.79 4.27
> > GBTOT -14.99 7.54
> >
> > The data here are not usable since the STD is bigger that the actual
> > value.
>
> Sorry, I don't see where the standard deviations are too high(?). I'm pretty
> sure that MMPBSA is reporting the mean and standard deviation of the
> distribution of values for each snapshot. If you want to estimate the
> standard error in the mean, you would have to divide the STD number by the
> square root of the number of independent samples you have. If the snapshots
> are widely separated in time (by more that a few tenths of a picosecond,
> generally), you can take the number of independent samples to be about equal
> to the number of snapshots.
>
> So, the estimated error in the PBTOT or GBTOT numbers is probably pretty
> small. So, if you had 100 snapshots, the estimated error in the mean value of
> GBTOT or PBTOT would be less than 1 kcal/mol.
>
> (All these are "statistical errors", of course, assuming that you indeed have
> a well equilibrated system. The actual errors, arising from deficiencies in
> the force field, and in the continuum solvent model itself, will generally be
> much larger than this.)
>
> ...dac
>
> -----------------------------------------------------------------------
> The AMBER Mail Reflector
> To post, send mail to amber_at_scripps.edu
> To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu
>
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu