> The data here are not usable since the STD is bigger that the actual
> value. I believe that the system is well equilibrated and wonder if
> there are other ways to improve. Thanks.
I am a bit confused here and do not exactly understand your concern here,
especially considering your previous e-mail where you argue that a
pressure of 0.6 bar (on average) should be closer to the target value
of 1.0 (despite large fluctuations). As you mention in the related
e-mail:
> Maybe I misunderstand the manu (p105) which says "...fluctuations in
> the instantaneous pressure on each step will appear to be large
> (several hundred bar), but the average value over many steps should be
> close to the target pressure..."
In my humble opinion, 0.6 bar is close to 1.0 bar, especially given a
fluctuation of +/- 200 bar.
Also, and in the same vein, the MM-PBSA averages can be meaningful
(although certainly not to kcal accuracy) despite large fluctuations. As
an example, you may consider my personal average position to the meter in
Utah as a function of time (over a year)-- it is roughly halfway between
home and the office, despite large fluctuations in the value (representing
travel to/from work/home and beyond). In general, one can have a very
meaningful average despite large fluctuations given *sufficient* sampling
of points. This is represented by the standard error rather than standard
deviation, i.e. stddev/#samples. In the earlier MM-PBSA papers, these
issues are discussed, including comparisons of values calculated over 200
frames say compared to 2000 frames of data.
Regarding the meaningfulness of your MM-PBSA averages, this is a debatable
and sometime heated issue (regardless of the standard deviations due to
such issues as estimates of entropy, solvation free energy, sampling,
conformational exchange, etc etc) and I do not know of a definitively
better way to get the insight you desire... You could try alternative /
complementary approach like linear interaction energy approaches, linear
response, free energy perturbation (if small change) or estimates of the
PMF of removing your ligand. Each of these has its strengths and
limitations. You could also try partitioning your trajectory (clustering)
into different substates (which may have different "energies" and lead to
larger standard deviations). Block averaging may also be useful (i.e.
doing the MM-PBSA over smaller blocks of the trajectory). The key in my
mind is whether the "energies" calculated give insight into your question
and are consistent with experiment/expectations. MM-PBSA is only an
approximate method, subject to real approximations, that may give insight
into (relative) energetics.
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
Received on Fri Apr 15 2005 - 00:53:00 PDT
