Adding to what Carlos remarked, "equilibration" vs. "production" depends on what you are looking at. There is no clear delineation, like a sudden green light that you've entered the production phase. This reminds me of our earlier Shao / Cheatham clustering paper. There are many ways to cluster - which one is best? There is not a best way unless that way helps partition the data as best fits your hypothesis and experiment. So try consensus of multiple approaches and know that partitioning is an art and not a rule - it is gray. It is not simply the case that our transitions between structures are 2-state; the transitions are smooth so where you make the cut in one state to another makes a difference in the result.
Knowing "equilibration" vs "production" depends on you knowing the answer up front. i.e. when are you out of range versus in-range for your measurable.
I can give a concrete example in terms of our DNA duplex simulations. We ran DNA simulations for years, thought them "production". Then we starting seeing base pair fraying on the termini. Were we "production" before we saw the base pair fraying? We thought so. Then fraying into two to three base pairs from the termini was observed. We again thought we were "converged", but what about low population Hoogsteen base pairs? Missed it. Now with longer simulations we see do internal base pair opening.and these low population Hoogsteen pairs. Would I say we are converged / production -- getting closer :-)
The key is not equilibration versus production, but are you sampling sufficiently to see the process of interest. If you are looking for a particular event (terminal base pair fraying and reforming, folding vs. unfolding, or drug binding / unbinding) normally you need to sample ~10 events (Nyquist) to accurately sample.
Our lab rule is run multiple sets of the simulations (with slightly different initial conditions aka ISEED, or different ion distributions, or different initial structures) and look for consensus. CPPTRAJ has great capabilities for combining and comparing independent runs and we puts scripts into the SI of our papers to help or see amberhub.chpc.utah.edu for tutorials (although I notice it is down again tonight, will fix).
tec3
________________________________________
From: Carlos Simmerling <carlos.simmerling.gmail.com>
Sent: Saturday, December 5, 2020 7:51:47 PM
To: AMBER Mailing List
Subject: Re: [AMBER] Identifying the equilibrated production region of a MD simulation trajectory
it's older, but this is a classic reference
https://pubmed.ncbi.nlm.nih.gov/12112673/
basically, the period when you are "equilibrated" depends on what you are
looking at.. energies etc plateau quickly, while reaching a converged
ensemble of course take much longer. many things are in between.
On Sat, Dec 5, 2020 at 2:41 AM Leena Aggarwal <leena.hrc.gmail.com> wrote:
> Dear AMBER Users
>
> Can anyone suggest a way for identifying the equilibrated production region
> in the MD simulation trajectory generated through AMBER?
> I want to do this using time series module of pymbar through
> detectEquilibration method. Is there any way to convert AMBER output to
> pymbar input?
>
> From
> Leena Aggarwal
> Department of Chemistry
> University of Delhi
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Sat Dec 05 2020 - 22:00:02 PST
