[gmx-users] g_wham Error bars

[chris.neale at utoronto.ca]

In addition to the suggestions made by Nicolas, you might try the wham  
program created by Alan Grossfield:  
http://membrane.urmc.rochester.edu/Software/WHAM/WHAM.html which is  
capable of doing a 'bootstrap' analysis of the error.

If you have the available computational resources, then I believe that  
Nicolas' suggestions are better than mine, but then again the  
bootstrapping method does not require that you basically throw away  
1/2 to 2/3 of your data into the error analysis. I make no comment on  
the utility of this bootstrapping method as I have always done  
something similar to what Nicolas suggested.

If you do go the route that was previously suggested, then you will  
want to also calculate some type of correlation time and ensure that  
your sampling is sig. larger than this, otherwise your error values  
may be underestimated without your knowledge.

Chris.


Hello,

As far as I know, it is not possible to directly calculate and error bar
with g_wham . At that point, you've got several options:

     1) You can run several times your umbrella sampling simulations,
     calculate several PMFs,  then calculate an average and a standard
     deviation
     2) if your simulations are split in several parts, you can calculate
     a PMF for each part and check their convergence over the time.
     3) if your PMF is supposed to be symmetrical, you can mirror it,
     calculate the difference between the PMF and its image, then used
     the difference/2 as a rough estimate of the error.

Hope that helps,
Nicolas

Raphael Alhadeff a écrit :
> Hello everyone,
>
> I wanted to ask if anyone knows how to compute standard deviation,  
> for the sake of adding error bars to an energy profile output using  
> g_wham (PMF using umbrella sampling in my case).
>
> Just to clarify, I am talking about the deviation of the energy  
> profile on ONE simulation (composed of 25 "windows" in my case, with  
> overlap but no repeats).
>
> Thanks in advance,
> Raphael
>
>
> -------------------