On Wed, Jun 22, 2016 at 6:43 AM, Neha Gandhi <n.gandhiau.gmail.com> wrote:
> There are number of techniques implemented in AMBER such as REMD, H-REMD,
> aMD, SMD and gaMD. In my view aMD methods and its enhancements (SMD, gaMD)
> are computationally less expensive since these techniques only require one
> simulation as compared to multiple replicas in REMD.
That is true, although they come with their own challenges
(reweighting etc). It is as true today as it always has been - there
is no "free lunch" :-).
> I saw paper by Daniel Roe on a very long simulation of nucleotide using
> combination of techniques such as aMD and H-REMD. In my view running long
> simulation using enhanced sampling techniques (for protein folding) doesn't
> guarantee convergence.
I think you may have missed the point of the paper. One of the main
issues we bring up in some of our recent papers (DOIs
10.1261/rna.051102.115, 10.1021/jp4125099, 10.1021/ct400862k, and
10.1021/jp400530e in particular) is that simply using an enhanced
sampling method doesn't mean you will obtain converged data within a
certain length of time; this is even explicitly stated in that
aMD/H-REMD paper you mention:
"While T-REMD has proven to be extremely useful in enhancing sampling,
its use does not guarantee convergence."
Which is why much of that paper (and the others) is devoted to
assessing to what extent multiple independent simulations have
converged.
> I was interested in knowing how can we assure
> whether a simulation has converged or not or whether a particular enhanced
> technique will do a better job over the other? Are running multiple
> simulations is one of the measure to look at statistically significant
> convergence?
I would say that running multiple (at least 2, ideally more)
independent simulations with different initial conditions and
different starting conformations is currently the best way to assess
convergence. We give present multiple ways of assessing convergence in
simulations in the references I give above, and there are many other
examples in the literature. We have had a lot of success with M-REMD,
but that doesn't mean it is a panacea for getting converged results.
You have to think about what the barriers to sampling in your system
might be and use a sampling technique that will best overcome them.
Also note that "convergence" is a blanket term and should be used
carefully. Different properties will converge at different rates. We
have typically looked at convergence of structural populations
obtained from clustering and principal components, but there are many
other things one can look at. Also, depending on what you want to do,
you may not need convergence. As with any problem, it depends on what
you want to do.
Hope this helps,
-Dan
>
>
> Any insight on the above matter is appreciated.
>
> Regards,
> Neha
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
--
-------------------------
Daniel R. Roe, PhD
Department of Medicinal Chemistry
University of Utah
30 South 2000 East, Room 307
Salt Lake City, UT 84112-5820
http://home.chpc.utah.edu/~cheatham/
(801) 587-9652
(801) 585-6208 (Fax)
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Jun 22 2016 - 06:30:02 PDT
