VMD-L Mailing List

From: Ryan McGreevy (ryanmcgreevy_at_ks.uiuc.edu)
Date: Wed Feb 28 2018 - 13:58:48 CST

It is hard to tell from the movie exactly what is happening, but if parts
of the structure are getting stuck where they shouldn't, in addition to
smoothing the map as you mentioned you tried, you could also try segmenting
the map prior to simulation. With a map segmented into different "parts"
(however you want to define them), you can run the MDFF simulation with
those multiple maps at the same time, each with their own corresponding
gridpdb file. This would allow for the forces from each sub-map to only
apply specifically to the atom selections you want, in this case for
example, making sure the stuck helix will only be affected by the forces
from the density it should fit into.

Because the gridforces underlying MDFF only "blindly" push atoms down the
local gradient, for more difficult modelling you will commonly have to use
schemes like these (domain restraints, smoothed and/or segmented maps,
initial finer-grained rigid body docking, IMDFF) to achieve the best
results.

On Wed, Feb 28, 2018 at 2:43 PM Rune Thomas Kidmose <rtk_at_mbg.au.dk> wrote:

> Hi Josh,
>
> Sorry for the late answer.
>
> Yes I have used COLORES for that many times and I really like it
> especially because it is so easy to run from cmd-line, however I have never
> heard of the other two programs.
>
> I was looking for a way of performing MDFF on an unaltered protein that
> have only been rigid fitted into a density, so no individual fitting of the
> domains etc. I have done this a couple of times now, but every time I try
> it on a target that needs large rotations it seems to fail.
>
> As mentioned earlier, I get that large domain rotations are not easy for
> MDFF to "solve", so I was basically just checking that I had not missed
> some setting or feature in MDFF that would make dealing with large
> rotations easier.
>
> As Ryan suggested I have also tried low-pass filtering the target densites
> (to like 15 or 20Å), and that actually helps some times, but large domain
> rotations are not resolved by this, at least not in my hands.
>
> However, it is not only large rotation I sometime have problems with, as
> an example I tried to MDFF the closed conformation of CorA channel (3jcf)
> into the low res density of one of the open conformations of the channel
> (3jch).
>
> I have made a video of the fitting and dropped it on youtube for those who
> might be interested.
>
> https://youtu.be/eXbK-KkEZr4
>
> Here you can see, I hope, how some of the long helices does not move into
> the "correct" density, which results in the whole domain, that needed to
> flip into the density, is stuck at a intermediate position.
>
> I have tried many different variations of MDFF parameters for this run,
> mainly playing with the G-scale (from 0.05 to 5) , numsteps (70.000 to
> 500.000), initial vs final temp, and I have never been able to get it
> right, though I get pretty close sometimes.
>
> Again I have not used domain restraints or individually fitted any of the
> domains, as they did in the original paper of the structures. I "just"
> wanted to do MDFF on a deposited model of the closed conformation rigid
> fitted into the open conformation, but I have not been able to do that.
> Maybe this is a good case for low-pass filtering the target density,
> despite it relativ low res to start with (7'ish Å).
>
> Cheers
>
> Rune
>
>
>
>
>
>
> ________________________________________
> Fra: Vermaas, Joshua <Joshua.Vermaas_at_nrel.gov>
> Sendt: 26. februar 2018 17:32
> Til: Rune Thomas Kidmose; Ryan McGreevy
> Cc: vmd-l_at_ks.uiuc.edu
> Emne: Re: Sv: vmd-l: MDFF rotation
>
> Hi Rune,
>
> Have you tried getting the domains prefit independently into the density,
> either with COLORES from Situs or newer EM-fitting programs like GEMfitter
> or BCL::EM-Fit? Then MDFF can fix up some of the small structural changes,
> letting the other programs get you in the right ballpark. It sounds like
> you are trying to take a structure that is misaligned to the initial
> density, which is going to be needlessly slow to converge rather than
> starting from a density-structure pair that are properly aligned.
>
> -Josh
>
> On 02/26/2018 08:15 AM, Rune Thomas Kidmose wrote:
>
> Hi Ryan,
>
>
> thx for getting back.
>
>
> Yes I have played a lot with the domain restraints feature as well, and
> for a few cases, as the one shown in the tutorial, it kinda works, but not
> for all.
>
>
> In general, I find the way you have to specify the domains using beta and
> occupancy values, to be a very slow'ish procedure for larger PDB files with
> many domains. But that is a different story I guess.
>
>
> Ruki
>
>
> ________________________________
> Fra: Ryan McGreevy <ryanmcgreevy_at_ks.uiuc.edu><mailto:
> ryanmcgreevy_at_ks.uiuc.edu>
> Sendt: 26. februar 2018 14:47
> Til: Rune Thomas Kidmose
> Cc: vmd-l_at_ks.uiuc.edu<mailto:vmd-l_at_ks.uiuc.edu>
> Emne: Re: vmd-l: MDFF rotation
>
> Have you tried using domain restraints (an example is in the MDFF
> tutorial)? This should help the domain fit together rather than the usual
> piecemeal approach.
>
> On Mon, Feb 26, 2018, 4:48 AM Rune Thomas Kidmose <rtk_at_mbg.au.dk<mailto:
> rtk_at_mbg.au.dk>> wrote:
>
> Hi all,
>
>
> I have been tinkering with MDFF for the past half year now, and I have
> accumulated a couple of questions, one of which is:
>
>
> Is it possible to somehow make MDFF handle large rotations of the initial
> model when fitting to the target density? It seems MDFF progresses quite
> linearly in its fitting, so is there a way to tell it or help it rotate
> e.g. a domain?
>
> I have come across a couple of cases where a domain clearly should be
> rotated significantly to fit the density, but MDFF just kept on pushing it
> into the density and then proceeded to fill the density by pulling in the
> domain. I am not thinking about iMDFF, where you manually interact with the
> model, which currently is the only way that worked for me.
>
>
> I acknowledge that rotation during MDFF it is not a trivial problem to
> solve but maybe I have missed some thing?
>
>
> Cheers
>
>
> Ruki
>
>
>
>