[gmx-users] Can g_wham support using different temperature for different windows?

[Justin A. Lemkul]

Jianguo Li wrote:
> Dear all,
>  
> Thank you all for your suggestions or comments to my problem. Now I am 
> planning to extend my simulations or using REMD in those bad windows to 
> get converged PMF.
> I have another question: if I extend the umbrellar simulation to 1 
> microsecond only in those problematic windows, while running shorter 
> simulation (e.g., 100 ns) in those windows far away from the 
> membrane. Does g_wham accpet using diffrent simulation time for 
> different windows?

Theoretically, but if you are going to discard any of the initial time (that is, 
equilibration in each window) by using the -b option, you'll have to make sure 
that you don't eliminate windows by doing so, i.e. 100 ns in some windows and 1 
us in others, using g_wham -b 200000 would completely neglect any windows with 
shorter time and basically make the PMF curve useless.

-Justin

> Thank you in again,
>  
> Cheers,
> Jianguo 
> 
> ------------------------------------------------------------------------
> *From:* XAvier Periole <x.periole at rug.nl>
> *To:* Discussion list for GROMACS users <gmx-users at gromacs.org>
> *Sent:* Wednesday, 23 February 2011 20:59:18
> *Subject:* Re: [gmx-users] Can g_wham support using different 
> temperature for different windows?
> 
> 
> On Feb 23, 2011, at 3:21 AM, Jianguo Li wrote:
> 
>>
>> Thank you  for the the useful information, XAvier.
>> My peptide is highly positively charged, 18 AA with +12 charges. Other 
>> of my group members told me their NMR experiment in water indicates 
>> the peptide conformation is very dynamics. Actually I also did peptide 
>> refolding using REMD in water, and I found it is flexible and has no 
>> stable structure in water, except some instantaneously helical 
>> structures. In addition, my peptide consists of two branches connected 
>> by unnatural peptide bond, so the backbone is discontinuous, and also 
>> because of the high charges, I assume the peptide doesn't form helcial 
>> structure in the negatively charged membrane. Therefore I didn't put 
>> any constraints in the peptide to keep the secondary structure of the 
>> peptide. I know there are assumptions in my model, but I have no other 
>> information to increase the accuracy of the model.  In fact, when I am 
>> doing REMD folding simulations using Gromos53a6 and CHARMM27 with 
>> cMap, I got different results. But the common thing is that both 
>> results seems to indicate the peptide is filexbile in water without 
>> stable secondary structure. Then I used MARTINI FF with flexible 
>> structure, just to find some general features.
>>
>> I will try your suggestion, doing REMD in those bad windows.
>>  
>> And the reference you mentioned is very useful, I will take a look at 
>> them :-)
>>
>> Another question: Suppose some other tools support using different 
>> temperatures in different windows, as you mentioned, if 500K is too 
>> high to have a significant contribution to the probability of 300K,  
>> can I do a series of simulation in a certain window with different 
>> temparatures (e.g. 300K, 350K, 400K,450K, 500K). In such cases, in 
>> each window, I need to do 5 simulations, which will be much cheaper 
>> than doing REMD in that window. 
> It would be computationally cheaper but this is assuming that you'd get 
> the info you are looking for within these simulations and again the 
> weight of the conformations from 400/450/500 K at 300 K is questionable. 
> Note also that the conformations sampled at high temperature with 
> position restrains on the lipids to avoid deformation will be difficult 
> to interpret! 
>>
>> Cheers
>> Jianguo
>>
>>
>> ------------------------------------------------------------------------
>> *From:* XAvier Periole <x.periole at rug.nl <mailto:x.periole at rug.nl>>
>> *To:* Discussion list for GROMACS users <gmx-users at gromacs.org 
>> <mailto:gmx-users at gromacs.org>>
>> *Sent:* Tuesday, 22 February 2011 21:18:12
>> *Subject:* Re: [gmx-users] Can g_wham support using different 
>> temperature for different windows?
>>
>>
>> A few notes:
>> - the original method (Kumar-JCC-1992) that inspired wham was actually 
>> developed to mix different temperature simulations. It is however not 
>> clear 
>> for the type of system you are simulating how much a 500K simulation 
>> would be useful to improve the sampling at 300 K or so. The reason is 
>> that the enthalpy difference between the two systems is so high that the 
>> probability that a conformation from a 500K simulation would contribute
>> to sampling at 300K is really low. It would much more efficient for 
>> systems 
>> with implicit solvent for which the energy of the system does not vary so 
>> much with the temperature. One could look at Chodera-JCTC-2007 
>> and ref therein for a few examples.
>> - I would think that a REMD simulation would be more useful. No need to
>> run 30 replicas to very hight temperature! A bilayer at 500K might get 
>> funny.
>>
>> - Martini force field for flexible regions of protein should not be 
>> trusted ... or
>> really interpreted with a lot of reserve. The "coil" definition is 
>> simply something
>> flexible with absolutely no guaranty that it could be representing 
>> some thing 
>> even close to reality, which we have only an approximate idea of what 
>> it is! 
>>
>> - A peptide in a bilayer has a very high chance to get into a helical 
>> conformation.
>> Do you think it is reasonable to keep it "flexible"? 
>>
>> - As noted by Justin and Chris, you definitely have a problem of 
>> convergence ... 
>> I am not sure how many "converged" examples of PMFs of peptide crossing a
>> bilayer are out in the literature (Justin?) but from our experience 
>> with Martini 
>> it does take an awful lot of time to really get convergence. For you 
>> system I 
>> would expect at least a microsecond for the windows where sampling is an 
>> issue. As an example, we saw significant differences on a PMF between two 
>> simple helices up to 8 us ... and no charges were involved.
>>
>> This might be a lot pessimistic but you should not get fooled by a CG 
>> model.
>> Martini is really good for a lot of things but other things should 
>> really but be
>> looked at carefully. 
>>
>> XAvier.
>>
>>  
> 

-- 
========================================

Justin A. Lemkul
Ph.D. Candidate
ICTAS Doctoral Scholar
MILES-IGERT Trainee
Department of Biochemistry
Virginia Tech
Blacksburg, VA
jalemkul[at]vt.edu | (540) 231-9080
http://www.bevanlab.biochem.vt.edu/Pages/Personal/justin

========================================