[gmx-users] Conformational sampling

chris.neale at utoronto.ca chris.neale at utoronto.ca
Tue Aug 10 01:56:45 CEST 2010


Simply applying available tools and seeing if the distribution looks  
okay is not a good plan. You should have a very well-defined idea of  
what you are trying to test and then pick a tool to get that done.

For example, perhaps you have a reference that provides the converged  
distribution of a dihedral and you want to see if your run reproduces  
this value. You would then use some tool, or combination of tools, to  
calculate the dihedral values and create a histogram of them.

Without an outside standard, it is certainly impossible to use a  
single run to prove that your sampling has converged. It is only  
possible to prove that it either (a) remains unconverged, or (b) that  
there is no evidence that it is unconverged.

In this case, if you really care about it, the best thing to do is to  
run a few separate simulations starting from different starting  
conformations (be sure to start from different dihedral basins). Then  
plot the time-average of some observables (e.g. the distribution of  
sampling of a given dihedral) and the values from different runs  
should "converge" to the same distribution if you run long enough --  
this is the true meaning of convergence, what many people do by way of  
things like block averaging is just a hack since one usually only has  
the resources to run a single trajectory.

And don't forget that there are more things to converge than just  
dihedrals. You could also look at pairwise combinations of dihedrals,  
etc. Note that I'm trying to keep the advice general here because I  
assume that you're eventually going to tackle something more complex  
than glycine.


> Well I guess, what I am trying to get at is, for my system I want to  
> make sure that 100ns has covered all the conformational changes  
> within the molecule, although I know there is not that much  
> conformational changes for glycine molecule, but I just wanted to  
> confirm. I did run g_angle command and I got the theta values for  
> different groups and the distribution looks okay according to me (I  
> compared the average angle values that I obtained from g_angle to  
> the actual angle values.

Then that's probably the best you can do for a molecule as simple as  
glycine.  I
would certainly think that rotation about such few bonds would happen within
even less time than 100 ns.


> Quoting "Justin A. Lemkul" <jalemkul at vt.edu>:
>> nishap.patel at utoronto.ca wrote:
>>> So is there a way I can test for convergence for my zwitterion for  
>>>  100ns run?
>> I'm not yet clear what you're assessing or how you define convergence.
>> In addition to what Chris said, you can look at dihedral transitions
>> with g_angle.  Surely there are a few dihedrals aside from standard
>> phi/psi, but I don't know what that's going to tell you.
>> -Justin
>>> Quoting "Justin A. Lemkul" <jalemkul at vt.edu>:
>>>> nishap.patel at utoronto.ca wrote:
>>>>> Okay so I tried to analyze the torsion using g_chi and g_rama  
>>>>> for   one glycine zwitterion in water. for g_rama I didn't see   
>>>>> anything  in xmgrace, and same for g_chi. I used this command  
>>>>> for  g_chi
>>>>> g_chi -f traj.xtc -s gly.gro -phi -psi
>>>>> When I run the command it says 1 residue with dihedrals found ,  
>>>>> 2   dihedrals found. But when I open the log file its empty and  
>>>>> so  are  the histo-phi/psiGLY.xvg plots. I tried using g_dih but  
>>>>> it  says:
>>>>> Found 0 phi-psi combinations
>>>> For a zwitterion, these torsions don't exist.  To measure phi, you need
>>>> at least C-N-CA-N, and for psi N-CA-C-N.  For a single zwitterion, you
>>>> have only N-CA-C.  You need at least a dipeptide.
>>>> -Justin
>>>>> I am not sure how to check for all torsion convergence for my    
>>>>> glycine zwitterion molecule. Am I missing something in the   
>>>>> command  line?
>>>>> -Nisha P
>>>>> Quoting chris.neale at utoronto.ca:
>>>>>> Nisha,
>>>>>> The approach is dictated by the goal. What do you want from this? and
>>>>>> why are you doing it? e.g. if you want to test the FF, then it is a
>>>>>> good idea to at least include US as a part of your strategy; if you
>>>>>> want to determine if 100 ns of equilibrium sampling is sufficient to
>>>>>> converge all torsions, then obviously you should not be doing US. In
>>>>>> any event, would start by running some equilbrium simulations and
>>>>>> analyzing the torsions with g_rama, g_chi, g_dih, etc. Be sure to read
>>>>>> through the -h output of each program as there are some nuances.
>>>>>> Chris.
>>>>>> -- original message --
>>>>>> Hello,
>>>>>>     I would like to do conformational sampling for my simulation of
>>>>>> one glycine in its zwitterionic form in water and obtain a PMF curve
>>>>>> to see if the system is equilibrated and that all possible torsions
>>>>>> are covered for my 100ns run. I am not sure how to approach this
>>>>>> issue. Is there a tutorial I can follow? Do I need to do umbrella
>>>>>> sampling and use WHAM to extract PMF?
>>>>>> I would appreciate some help!
>>>>>> Thanks.
>>>>>> Nisha P.

More information about the gromacs.org_gmx-users mailing list