[gmx-users] Asymmetry in homo dimer simulation

[Justin Lemkul]

On 12/6/12 5:26 AM, Erik Marklund wrote:
>
> 5 dec 2012 kl. 17.26 skrev Justin Lemkul:
>
>>
>>
>> On 12/5/12 11:21 AM, Kavyashree M wrote:
>>> Sir,
>>>
>>> Thank you for your suggestions. I decided the cutoff based on
>>> RMSD convergence. I will calculate at different time intervals.
>>> Running multiple simulation is definitely the best suggestion but
>>> due to time and machine constraint it would be difficult. Instead
>>> I have two mesophilic simulations. But Is there any other way
>>> by which I can prove this point?
>>>
>>
>> Not using single trajectories.  Your job is to convince reviewers that your work is sound.  A single trajectory is not convincing, at least to any reviewer that does his or her homework :)
>>
>
> I don't think that the use of single trajectories is necessarily "wrong", as long as they are sufficiently long. It's usually the amount of sampling that is the crucial point, no?
>

Well, long simulations have issues too - 
http://dx.doi.org/10.1016/j.bpj.2010.04.062.

That said, there's probably not one single answer here.  I am generally leery of 
anyone trying to make arguments based on a single trajectory, especially one 
that is relatively short by modern standards.  Indeed, the simulation needs to 
be sufficiently long to potentially observe the phenomenon of interest, and that 
simulation should be converged and of adequate length to collect reasonable 
statistics.  I think, from a practical standpoint, several simulations of 
intermediate length will give you better insight than a single simulation of any 
length.  Certainly in my work on amyloid peptides, multiple trajectories of 
considerable length are prerequisite; most other papers get (rightly) thrown out 
immediately.  In the case of well-folded proteins whose behavior is more 
predictable, you can probably get more information from a single high-quality 
trajectory.

In the present case, where a simulation of modest length is not agreeing with 
(presumably) experimental data, I think sampling or convergence is a major 
concern.  Had the trajectory reproduced the expected behavior, it's less of an 
issue.

-Justin

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

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

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