Re: Multi-dimensional PMF

From: Jeff Comer (jeffcomer_at_gmail.com)
Date: Fri Jul 27 2018 - 18:39:37 CDT

Dear Kelly,

If you are interested in the free energy along all five dimensions, you can
usually do 1D ABF (which often helps with sampling along all degrees of
freedom) and reweight to obtain PMFs along other directions. In this paper
( http://doi.org/10.1021/acs.jcim.7b00521 ), we generate a 3D PMF from a
trajectory where ABF was applied along a single dimension. See equation 1.
To do this, you should store a lot of frames or use Colvars to get good

In any case, I do 2D and 3D ABF pretty regularly. In this paper (
http://doi.org/10.1021/acs.jpcb.7b01130 ), we use a volumetric
representation for the 3D PMF (see Figure 3A), although it's not that easy
to look at. You can either integrate these PMFs into lower dimensions or
take slices out of them. Slices aligned with the collective variable axes
can be done easily with awk, Matlab, etc. We made 2D heat maps using slices
of the 3D PMF (see Figure 3B,C,D). You can also integrate 2D and 3D PMFs to
obtain lower dimensional PMFs (Figure 3F). If the inclusion of the extra
dimensions is to get better sampling along the primary dimension, then all
you really care about is this 1D PMF anyway. In the limit where sampling is
good along the other dimensions in the absence of multidimensional ABF, 1D
ABF and 3D ABF integrated down to one dimension give the same results.

One warning: 3D ABF requires a lot of sampling and larger "widths" than
you'd use for 1D ABF (to avoid having millions of bins). 4D might be
possible...

Have fun,
Jeff

–––––––––––––––––––––––––––––––––––———————
Jeffrey Comer, PhD
Assistant Professor
Institute of Computational Comparative Medicine
Nanotechnology Innovation Center of Kansas State
Kansas State University
Office: P-213 Mosier Hall
Phone: 785-532-6311
Website: http://jeffcomer.us

On Fri, Jul 27, 2018 at 2:43 PM, Vermaas, Joshua <Joshua.Vermaas_at_nrel.gov>
wrote:

> No, I mean figure out what parts of the variables are actually relevant to
> the transition you are trying to describe and make it 1-D. The pathCV is
> just useful if you can't entirely get a handle on what *specifically* is
> changing, but you know that something is, since RMSD to a target hides a
> multitude of sins. Multidimensional sampling, while technically doable,
> is expensive to do properly, since the amount of sampling required for
> convergence is polynomial in the number of dimensions. So 1D is ideal, 2D
> is doable, and anything beyond is unheard of.
>
> If it were me, I'd use a regular distanceZ on the parts away from the
> histidines, and just focus sampling on the transition like crazy. This
> sounds like a selectivity filter for the channel, where the ions tend to
> dehydrate quite substantially, and you get a hard-to-sample region of your
> PMF. How long did you run your 1D stuff again? Di Maio et a.
> (10.1371/journal.pone.0140258) ran for 20ns a window, and while their
> sodium PMF closes (equal free energies in solution), their chloride PMF
> does not, and seems to "remember" being dragged across the membrane to
> start the simulation.
>
> -Josh
>
>
>
> On 2018-07-27 12:07:57-06:00 McGuire, Kelly wrote:
>
> Thanks Josh, just to be clear when you say project down from 5D to 1D you
> do mean a vector projection from the 5D output values to a 1D vector?
>
>
> *Kelly L. McGuire*
>
> *PhD Scholar*
>
> *Biophysics*
>
> *Department of Physiology and Developmental Biology*
>
> *Brigham Young University*
>
> *LSB 3050*
>
> *Provo, UT 84602*
> ------------------------------
> *From:* Vermaas, Joshua <Joshua.Vermaas_at_nrel.gov>
> *Sent:* Friday, July 27, 2018 10:27:45 AM
> *To:* McGuire, Kelly; jerome.henin_at_ibpc.fr
> *Cc:* namd-l_at_ks.uiuc.edu
> *Subject:* RE: namd-l: Multi-dimensional PMF
> Hi Kelly,
>
> Figure out the appropriate 1-D pmf so that other collective variables are
> no longer corrupting. This might involve sampling a higher dimensional
> space, and then projecting down from that higher dimension to make a 1-D
> reaction coordinate. An example of this is the "pathCV" example provided in
> the colvars source (https://github.com/Colvars/
> colvars/blob/master/examples/10_pathCV.namd
> which basically defines a pathway based on distinct states, and
> interpolates between them. I've found the paper cited within to be very
> informative.
>
> -Josh
>
>
>
> On 2018-07-27 09:57:08-06:00 owner-namd-l_at_ks.uiuc.edu wrote:
>
> Ah ok I see. In your experience, when other degrees of freedom in your
> system are corrupting your 1-D PMF, what is your process to deal with them?
> If it requires higher dimensions because you have to use more colvars, do
> you stay away from those situations? It was suggested to me that I bias the
> other orthogonal degrees of freedom, which is how I ended up at 5D.
>
>
> *Kelly L. McGuire*
>
> *PhD Scholar*
>
> *Biophysics*
>
> *Department of Physiology and Developmental Biology*
>
> *Brigham Young University*
>
> *LSB 3050*
>
> *Provo, UT 84602*
> ------------------------------
> *From:* Jérôme Hénin <jerome.henin_at_ibpc.fr>
> *Sent:* Friday, July 27, 2018 9:16:44 AM
> *To:* McGuire, Kelly
> *Cc:* Namd Mailing List
> *Subject:* Re: namd-l: Multi-dimensional PMF
> What I meant between the lines is, don't do 5D PMFs.
> There are no ready-made tools to do the projection - if you get into this
> you'll have to be ready to write your own tools, and know the corresponding
> theory.
> Jerome
>
> On Fri, 27 Jul 2018 at 16:59, McGuire, Kelly <mcg05004_at_byui.edu> wrote:
>
> It's been a while since I've dealt with projections. Any suggestions on
> programs or the process you use to do that?
>
> Kelly L. McGuire
> PhD Scholar
> Biophysics
> Department of Physiology and Developmental Biology
> Brigham Young University
> LSB 3050
> Provo, UT 84602
>
> ------------------------------
> *From:* Jérôme Hénin <jerome.henin_at_ibpc.fr>
> *Sent:* Friday, July 27, 2018 1:45:20 AM
> *To:* Namd Mailing List; McGuire, Kelly
> *Subject:* Re: namd-l: Multi-dimensional PMF
> Hi Kelly,
> the only way to plot 5D data is to project it into lower-dimension spaces--00000000000004a9ee057203a046--

This archive was generated by hypermail 2.1.6 : Mon Dec 31 2018 - 23:21:19 CST