Re: Re: Thermodynamic integration script - where is it?

From: Floris Buelens (floris_buelens_at_yahoo.com)
Date: Tue Jun 16 2009 - 02:37:44 CDT

Hi Lela,

> Are there some documents on the web (or plans to put them on the web in
> near future) where we can check more information about the method
> results, recommendations, and method validity? We would first like to
> be confident about general settings compatible with TI method, such as
> which ensemble is appropriate, the recommended Langevin damping, etc.

I think it's best to refer to published reports for information like this - TI is a new feature for NAMD but has been widely applied elsewhere. One rule of thumb is that any settings appropriate for the FEP functionality in NAMD should be fine for TI as well.

> Also, we are considering testing the lengths of the trajectories for
the
> equilibration and subsequent ensemble average collection. Upon
testing
> the lengths of the trajectories (for example collecting
ensemble averages
> for 5,000 versus 10,000 steps for each lambda),
should obtaining the
> reproducible results for free energy differences
on the same system
> tell us that we are doing things right?

Two categories of error can be described for TI. First there will be a statistical error, which arises from variance in collected data (dU/dl). If you see reproducible results between runs this is a sign that the statistical error is acceptable. Secondly, there will be a systematic error associated with the estimation of the integral; in my opinion the best way to assess this error is to convince yourself by visual inspection of the dU/dl curve that the transformation is sampled adequately smoothly.
This article and references therein are a useful starting point for these issues: Ytreberg et al. Comparison of free energy methods for molecular systems. The Journal of Chemical Physics (2006) vol. 125 (18) pp. 184114.

> In addition, I have two specific questions about NAMD_ti.pl. On page
127
> of the NAMD user guide, there is a plot of log(dE/dl) versus lambda
for
> checking the adequacy of lambda sampling. Our question is which
dE/dl
> is used for this plot? The full output of NAMD_ti.pl script
reports these
> derivatives separately for vdW and Coulombic
interactions.

That plot was produced using data from the vdW component for the insertion of a small molecule into a condensed phase system.

> The second question is about the warning message we get: "Warning: one
> or more integrands were worryingly large (>30); run with
--fulloutput and
> check consistency." In the plot on the page 127 of
user guide, the
> integrands are of the 10^12 order of magnitude, and we
get integrands
> up to 10^5 for the vdW dE/dl. I am not sure if we should
be worried.

First of all the use of the word 'integrand' in this warning is incorrect (a mistake on my part) - what this warning actually refers to is that the area under the curve for one or more sections of the integral (e.g. from one lambda value to the next) is on the large side. dU/dl values can be far larger, but if the lambda values are spaced close together the area under the curve will remain small.
In short, you probably should be worried; this means that you're effectively measuring a free energy difference of 30 kcal/mol going from one lambda value to the next, which is huge. I would suggest looking at the early parts of your vdW insertion (using --fulloutput) and identifying regions in which finer sampling (more lambda values, spaced closer together) may be necessary.

There will be a tutorial document published for TI (along the same lines as the FEP ones that are currently available); I'll be sure to work through some of these issues. Please don't hesitate to email with more questions as it's good to see what kind of issues new users are confronted with.
Best wishes,

Floris

________________________________
From: Lela Vukovic <lvukov1_at_gmail.com>
To: Floris Buelens <floris_buelens_at_yahoo.com>
Sent: Monday, 15 June, 2009 19:08:19
Subject: Re: namd-l: Re: Thermodynamic integration script - where is it?

Hello Floris,

we have started using TI method, and we have several questions. Are there some documents on the web (or plans to put them on the web in near future) where we can check more information about the method results, recommendations, and method validity? We would first like to be confident about general settings compatible with TI method, such as which ensemble is appropriate, the recommended Langevin damping, etc. Also, we are considering testing the lengths of the trajectories for the equilibration and subsequent ensemble average collection. Upon testing the lengths of the trajectories (for example collecting ensemble averages for 5,000 versus 10,000 steps for each lambda), should obtaining the reproducible results for free energy differences on the same system tell us that we are doing things right?

In addition, I have two specific questions about NAMD_ti.pl. On page 127 of the NAMD user guide, there is a plot of log(dE/dl) versus lambda for checking the adequacy of lambda sampling. Our question is which dE/dl is used for this plot? The full output of NAMD_ti.pl script reports these derivatives separately for vdW and Coulombic interactions.

The second question is about the warning message we get: "Warning: one or more integrands were worryingly large (>30); run with --fulloutput and check consistency." In the plot on the page 127 of user guide, the integrands are of the 10^12 order of magnitude, and we get integrands up to 10^5 for the vdW dE/dl. I am not sure if we should be worried.

I would greatly appreciate your help in this, and information about if and when some additional documents about this method will be placed on the NAMD website.

Thank you,
Lela Vukovic

2009/6/2 Lela Vukovic <lvukov1_at_gmail.com>

Hi Floris,
thank you very much!
Lela

2009/6/2 Floris Buelens <floris_buelens_at_yahoo.com>

Hi Lela,

The script hasn't made it onto the web server yet, I've attached it. Please don't hesitate to email if you have any questions or problems as it hasn't been used very widely yet and I've made a couple of changes recently.
Best wishes,

Floris

________________________________
From: Lela Vukovic <lvukov1_at_gmail.com>
To: namd-l_at_ks.uiuc.edu
Sent: Monday, 1 June, 2009 18:52:38
Subject: namd-l: Re: Thermodynamic integration script - where is it?

Hello,
I am testing the thermodynamic integration within MD, but I am missing a necessary script to finalize the simulations:
in the user's guide, it is stated that there is "NAMD_ti script, available from http://www.ks.uiuc.edu/Research/namd/utilities/ "

However,
I am unable to locate it. Does anybody know a direct link to the script?
Thank you in advance,
Lela

      

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:52:55 CST