Re: ABF with the RMSD colvar

From: Jérôme Hénin (jerome.henin_at_ibpc.fr)
Date: Mon Sep 22 2014 - 08:37:00 CDT

On Sep 22, 2014 1:22 PM, "George Patargias" <gpat_at_bioacademy.gr> wrote:
>
> Hello Jerome
>
> > Try plotting the PMF calculated by ABF (even unconverged) and -RT
> >log(samples per bin), and see how those two compare. The log sampling
> >should be significantly flatter than the PMF.
>
> Apart from the inf values for the bins with 0 samples, the -RTlog(samples
> per bin) plot looks more rugged than the PMF. With "spiky" minima at 2,4,6
> and 8 Ang RMSD values.
>
> > The accumulation at one side could depend from non-equilibrium effects,
> >and depend on the history of the system. Was nonequilibrium pulling
> >involved at some point? Did it have time to relax?
>
> I run a 10 ns Targeted MD (using the the same RMSD colvar but with
> "harmonic", instead of "ABF" bias). Inspecting the .colvars.traj output
> file from this TMD, I extracted snapshots with an RMSD value between the
> upper and the lower boundary of every window. For example, for the window
> 8 - 10 Ang, I chose the snapshot with an RMSD value of 9 Ang
>
> It did not have time to relax, i.e there was no standard MD
> (equilibration) before the ABF sim in every window.

That's probably a major cause of your problems. ABF remains close to
equilibrium, if all goes well, but if you start it away from equilibrium,
that will cause trouble. Essentially, it will record non-equilibrium
forces, and those will skew the biasing force and yield uneven sampling.

Jerome

> >What happens if you start a free simulation instead of ABF?
>
> From a short free MD that I run on the snapshot of the first ABF window,
> the RMSD is stabilized after ~1.2 ns.
>
> Thanks again.
> George
>
>
> > Best,
> > Jerome
> > On 19 September 2014 17:12, George Patargias <gpat_at_bioacademy.gr> wrote:
> >> Hello Jerome,
> >> Thanks for your reply.
> >> I set up this calculation in 5 x 2 Ang windows with 2 ns ABF/MD in
> each.
> >> From the .count files, I noticed that, in all windows, the bins that
> correspond to higher RMSD values are never visited, i.e the number of
> collected samples is 0. Actually the first bin (towards the lower
> boundary) has 7-8 times more samples than the rest.
> >> This is probably why the RMSD values (from the .traj files) fluctuate
> mostly around the lower RMSD value of every window.
> >> Can this be caused by a too low or too high Nsamples value? I have set
> it
> >> to 1000. Or maybe too low force constant for lower and upper boundary
> values - I have set it to 100 Kcal/mol/Ang^2.
> >> I copy/paste my colvars file used for the first window. Thanks again!
> George
> >> colvar {
> >> name RMSD
> >> width 0.1 #Check again
> >> lowerboundary 8.0
> >> upperboundary 10.0
> >> lowerwallconstant 100.0
> >> upperwallconstant 100.0
> >> outputAppliedForce on
> >> rmsd {
> >> atoms {
> >> atomsfile equil2_Arp2Arp3CA.pdb
> >> atomsCol B
> >> atomsColValue 100.00
> >> }
> >> refPositionsFile Arp23_ac_min2_Arp2Arp3.pdb
> >> refPositionsCol B
> >> refPositionsColValue 100.00
> >> }
> >> }
> >> abf { # Define an ABF bias on RMSD colvar
> >> colvars RMSD
> >> fullSamples 1000
> >> > Hi George,
> >> > Yes, it's definitely possible to run this in several windows. Be
> aware that RMSD coordinates have strong Jacobian terms - in short,
> >> they
> >> > have a strong tendency to drift to larger values in the absence of
> any
> >> physical interactions. The ABF bias will compensate for that
> >> automatically
> >> > by applying significant negative biasing forces, especially at lower
> >> values
> >> > of the coordinate. In any case, there is a singularity around RMSD =
> >> 0,
> >> where the PMF goes to infinity for purely geometric reasons.
> >> > This might makes it a little less practical to work with RMSD
> >> coordinates
> >> > compared with more regular functions. At some point I experimented
> >> with
> >> a
> >> > logMSD coordinate which I expected to behave better, and wasn't
> >> totally
> >> satisfied with the results.
> >> > Best,
> >> > Jerome
> >> > On 17 September 2014 13:43, George Patargias <gpat_at_bioacademy.gr>
> >> wrote:
> >> >> Hello,
> >> >> As I have posted before, I am trying to study the conformational
> >> change
> >> of
> >> >> a protein using the RMSD colvar.
> >> >> From a 10 ns TMD simulation (harmonic type of bias in the colvar
> >> configuration file), I have a pretty good idea of the conformational
> pathway.
> >> >> I would like now to extract a PMF for this conformational change
> >> using
> >> the
> >> >> ABF method.
> >> >> From some test ABF/MD runs, the RMSD value drops from an initial
> >> value
> >> of
> >> >> 9.8 Ang (set as upperboundary) to 0.8 Ang (set as lowerboundary)
> >> within
> >> 400 ps and stays close to this value for the rest of the simulation. I
> also tried to run a short ABF sim with 9.8 Ang (upperboundary) to 6.8 Ang
> >> (lowerboundary) and with 100 kcal/mol/Ang^2 lower and upper wall
> constant
> >> but the RMSD still becomes lower than the lowerboundary value. Is it
> plausible to run ABF sims, let's say, for 3 windows (9.8 - 6.8, 6.8
> >> >> - 3.8 and 3.8 - 0.8) with an initial RMSD value somewhere *between*
> >> the
> >> range values of every window (like in the AmtB transporter tutorial)?
> Many
> >> thanks in advance.
> >> >> George
> >> >> > The averaging of the ensemble of states of a single simulation is
> >> done
> >> >> by
> >> >> > using the accumulated work flag itself (forces working in favor of
> >> the
> >> >> transformation or against it will be averaged simply by integrating
> >> both).
> >> >> > To average between multiple simulations, use the well known
> >> >> Jarzynski's
> >> >> > formula.
> >> >> > You can define the same exact colvar but apply different methods
> to
> >> it
> >> >> and
> >> >> > thus obtain different results, which should be equivalent in the
> >> limit
> >> >> of
> >> >> > very long simulation time.
> >> >> > You mentioned a 7-subunits protein, i.e. a very complex system,
> for
> >> >> which
> >> >> > you should anticipate that to obtain a reliable PMF won't be easy.
> >> >> Doing
> >> >> > preliminary tests such as a steered MD (aka a targeted MD in this
> >> >> case)
> >> >> to
> >> >> > get an idea of the transformation pathway can be a good idea.
> Then
> >> >> when
> >> >> you know a bit about the transformation, use whichever free energy
> >> calculation method you think most appropriate.
> >> >> > Giacomo
> >> >> >> Best wishes
> >> >> >> George
> >> >> >> > On Wed, Apr 16, 2014 at 6:14 AM, George Patargias
> >> >> >> > <gpat_at_bioacademy.gr>wrote:
> >> >> >> >> Hi Giacomo,
> >> >> >> >> Sorry for the hassle; just one more question on this
> particular
> >> >> ABF
> >> >> >> calculation.
> >> >> >> >> If I want to study the conformational transition A --> B and
> >> use
> >> >> the
> >> >> >> structure of B as a reference for the RMSD colvar, is the ABF
> bias
> >> >> going
> >> >> >> to "drive" the RMSD of A with respect to B from the upperboundary
> >> >> value
> >> >> (that I will calculate by superimposing A and B) to the
> >> >> >> >> lowerboundary
> >> >> >> >> value (a small one, like 0.1)?
> >> >> >> >> George
> >> >> >> >> > Yes, avoid using wrapAll in this case. Non covalently
> linked
> >> >> >> protein
> >> >> >> >> fragments would be wrapped individually, and mess up the
> >> >> calculation
> >> >> >> of
> >> >> >> the
> >> >> >> >> > RMSD.
> >> >> >> >> > Giacomo
> >> >> >> >> > On Tue, Apr 15, 2014 at 6:22 AM, George Patargias
> >> >> >> >> > <gpat_at_bioacademy.gr>wrote:
> >> >> >> >> >> Hello,
> >> >> >> >> >> I am trying to set up an ABF calculation using the RMSD
> >> colvar.
> >> >> >> The
> >> >> >> >> atom
> >> >> >> >> >> block of the colvar configuration file contains all the
> >> C-alpha
> >> >> >> atoms
> >> >> >> >> of
> >> >> >> >> >> a
> >> >> >> >> >> protein complex that consists of 7 (non covalently linked)
> >> >> >> subunits.
> >> >> >> >> I
> >> >> >> >> am trying to decide whether I need to exclude the wrapAll
> >> option
> >> >> (and
> >> >> >> use only wrapWater) on the basis of the recommendations found
> here
> >>
http://www.ks.uiuc.edu/Research/namd/2.9/ug/node55.html#SECTION000132410000000000000
> I would really appreciate any tips on this
> >> >> >> >> >> Thanks!
> >> >> >> >> >> Dr. George Patargias
> >> >> >> >> >> Postdoctoral Research Fellow
> >> >> >> >> >> Biomedical Research Foundation
> >> >> >> >> >> Academy of Athens
> >> >> >> >> >> 4, Soranou Ephessiou
> >> >> >> >> >> 115 27
> >> >> >> >> >> Athens
> >> >> >> >> >> Greece
> >> >> >> >> >> Office: +302106597568
> >> >> >> >> Dr. George Patargias
> >> >> >> >> Postdoctoral Research Fellow
> >> >> >> >> Biomedical Research Foundation
> >> >> >> >> Academy of Athens
> >> >> >> >> 4, Soranou Ephessiou
> >> >> >> >> 115 27
> >> >> >> >> Athens
> >> >> >> >> Greece
> >> >> >> >> Office: +302106597568
> >> >> >> Dr. George Patargias
> >> >> >> Postdoctoral Research Fellow
> >> >> >> Biomedical Research Foundation
> >> >> >> Academy of Athens
> >> >> >> 4, Soranou Ephessiou
> >> >> >> 115 27
> >> >> >> Athens
> >> >> >> Greece
> >> >> >> Office: +302106597568
> >> >> Dr. George Patargias
> >> >> Postdoctoral Research Fellow
> >> >> Biomedical Research Foundation
> >> >> Academy of Athens
> >> >> 4, Soranou Ephessiou
> >> >> 115 27
> >> >> Athens
> >> >> Greece
> >> >> Office: +302106597568
> >> Dr. George Patargias
> >> Postdoctoral Research Fellow
> >> Biomedical Research Foundation
> >> Academy of Athens
> >> 4, Soranou Ephessiou
> >> 115 27
> >> Athens
> >> Greece
> >> Office: +302106597568
>
>
> Dr. George Patargias
> Postdoctoral Research Fellow
> Biomedical Research Foundation
> Academy of Athens
> 4, Soranou Ephessiou
> 115 27
> Athens
> Greece
>
> Office: +302106597568
>
>
>
>
>
>
>
>
>
>
>
>

This archive was generated by hypermail 2.1.6 : Wed Dec 31 2014 - 23:22:52 CST