On Wed, 2014-05-28 at 13:36 +0000, Aronica, Pietro wrote:
> As a follow-up, I've been using the older version of MMPBSA (2.7) as a replacement for now. It gives results that are comparable to those of the newer version. The magnitude and the error of the deltas for both native and mutant are very similar, as well as the magnitude of the delta delta. However, the error of the delta delta is huge, ten times what it is in the newer version.
> To make an example, this is the MMPBSA calculation on the same trajectory.
>
> Newer version:
> Native: -53.8694 +/- 7.2906
> Mutant: -49.1516 +/- 7.2006
> Delta delta: -4.7178 +/- 1.0669
>
> Older version:
> Native: -53.8586 +/- 7.2821
> Mutant: -49.1035 +/- 7.1908
> Delta delta: -4.7551 +/- 10.2341
>
> Why is the error so large? Is it calculated in a different way?
Yes. The older version computed the standard deviation (and std. err.
of the mean) by standard uncorrelated error propagation (square root of
the sum of the variances).
The new version calculates the Delta-delta for every frame and computes
the standard deviation of that population. This approach fully accounts
for the correlation between the data sets (which is why the standard
deviation is so much lower).
The same is true for decomposition results (old versions used error
propagation to compute uncertainties and fluctuations in the DELTA
values whereas the newer versions construct DELTA populations and
compute fluctuations and uncertainties directly from there).
HTH,
Jason
P.S. I'm still trying to find the root cause of the alanine scanning
issue...
--
Jason M. Swails
BioMaPS,
Rutgers University
Postdoctoral Researcher
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed May 28 2014 - 07:00:04 PDT
