From: Giacomo Fiorin (giacomo.fiorin_at_gmail.com)
Date: Wed Jun 26 2019 - 17:20:57 CDT
That's a tricky question, because estimating the standard error requires
that you sampled the underlying distribution in an unbiased manner. If you
use WHAM or MBAR (two widely used methods to unbias umbrella-sampling
distributions), most implementations of each provide an estimate of the
error based on certain assumptions.
But if your choice of collective variable(s) fails to capture all the
relevant degrees of freedom, all those assumptions fail because of
histeresis, and the difference between the true and computed PMFs (i.e. the
*real* error bars) would be much larger. It would help if you computed a
second set of runs initialized differently (e.g. by a backward instead of
forward sweep) and compared the two PMFs. If they are consistent, you can
combine the sampling to get a single one.
Also, when you present your results to an audience that is not just
computational (as you should) you need to point out that the errors in the
physical model are never captured, even with perfect statistical sampling.
The amount of this error depends on the force field and on how you prepared
Lastly, these are very generic statements and you need to look more
specifically at literature papers on problems similar to yours, which are
probably not few.
On Wed, Jun 26, 2019 at 6:03 PM Mahdi Mousaei <mahdi.mousaei_at_ucalgary.ca>
> Hello to all,
> I used NAMD for Umbrella Sampling and I have 120 Windows which are
> simulated for ~100ns. I have the PMF profile now and want to add the error
> bars to my PMF. It would be great if anyone could help me in this regard.
> Best Wishes,
-- Giacomo Fiorin Associate Professor of Research, Temple University, Philadelphia, PA Research collaborator, National Institutes of Health, Bethesda, MD http://goo.gl/Q3TBQU https://github.com/giacomofiorin
This archive was generated by hypermail 2.1.6 : Wed Sep 18 2019 - 23:21:04 CDT