[gmx-users] a residue move in extremely large scale in MD

[Justin Lemkul]

On 8/11/12 7:25 PM, Acoot Brett wrote:
> Dear Albert,
>
>
> I hope I can support my MD results with some bioinformatics data. For
> example, it indicates the highly movavle residue I mentioned in my previous
> e-mail is not conserved at all, and it is only specific that structure.
>
> For your 10X with different seed simulation, does it mean the production time
> scale changes to 100 ns, or does it mean I run 10 rounds of 10 ns MD to
> vierify the results? For GROMACS, the seed is choosen automatically, do we
> have the method to choose it manually?
>

There are different schools of thought here.  Since MD simulations are highly 
prone to getting stuck in local energy minima, a single simulation that is very 
long is not necessarily better than several simulations of intermediate length. 
  There are numerous methods for enhancing sampling.

One can set different starting velocities manually by simply setting (very) 
different values of gen_seed at the outset of any simulation.  The default value 
of -1 takes the seed from the PID of executing grompp, which means you can get 
different velocities simply by running the command several times.

> For the converge you mentioned, I agree for some simulations 10 ns is not
> enough, but I think within 10 ns if what you want to determine has converged,
> that would be fine. For converge, which parameters do you use to decide
> whether it has converged?
>

Convergence is a detailed topic that one can write several book chapters about. 
  Existing literature covers most of the conventional approaches.  Read enough 
MD literature and you'll have a fair idea of how to properly assess it.

> There are some Journals accept computational biology research results without
> the support of biochemistry data, can you suggest these Journals to me?
>

In my experience, reviewers are getting more and more skeptical, so pure 
theoretical demonstrations that that have no demonstrated connection to 
experimental work are not very strong, in general.  You have to have some 
evidence to suggest that your prediction has any relevance to an actual 
biological system.

-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

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