Menu
▾
▴
Re: [scipion-users] HEMNMA visualization and nma-align efficiency
|
From: Slavica J. <sla...@up...> - 2021-11-27 12:12:00
|
Dear Lizelle,
Thanks for your email and for using HEMNMA (and, more generally,
ContinuousFlex plugin of Scipion). Here are a few hints and
recommendations. If you need more help, let me know by emailing me directly.
*1) Visualization of normal-mode animations:*
In Scipion, VMD is the only option for visualizing normal-mode animations.
However, you can download the animations saved on your distant computer and
visualize them on your local computer using VMD if you have it installed. I
will give you some instructions here on what to do in both cases.
For playing animations with VMD in Scipion:
----------------------------------------------------------
The problem with VMD on your distant computer may be either related to (1)
a slow connection from your local computer (if you see that VMD is trying
to open the animation but it is taking too long and the animation cannot
open at the end) or to (2) a missing path to the VMD binary in the
configuration file of Scipion (check ~/scipion3/config/scipion.conf and add
the path to VMD binary that you will get using "which vmd" on a command
line of a terminal; for instance, if this command is giving you
"/usr/local/bin/vmd", it means that the binary is in "/usr/local/bin" in
which case you would add "VMD_HOME = /usr/local/bin" to scipion.conf, but
without quote marks).
For downloading animations from your distant computer and playing them with
VMD on your local computer (Scipion is not required for this):
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
All calculated animations are saved on the following path on your distant
computer (assuming ScipionUserData is in your home directory):
~/ScipionUserData/projects/NAME_OF_YOUR_PROJECT/Runs/XXXXXX_FlexProtNMA/extra/animations
Note that "NAME_OF_YOUR_PROJECT" should be replaced by the name you gave to
your project when creating it in Scipion. Aslo, "XXXXXX" should be replaced
by the run number that was automatically assigned by Scipion when you
started this NMA run.
For each normal mode, you will see two files: *.vmd and *.pdb (only the
normal modes higher than mode 7 are animated, as the first 6 modes should
not be used with HEMNMA). The file with the extension "pdb" (*.pdb) is the
animated PDB that VMD knows to play even without a dedicated vmd script
(the file with "vmd" extension, i.e., *.vmd). You can download all these
animations and play them on your local computer if you have VMD installed
on it. Load one *.pdb using "New Molecule" in "File" menu of VMD, press the
play button of VMD and the animation will start. If you wish a nicer
graphical representation, you may use the associated *.vmd script for this,
but you should replace the first line of this script by the path on which
the animation is located locally on your computer. For instance, you may
replace the first line of animated_mode_007.vmd by "mol new
./animated_mode_007.pdb", without quote marks, if you wish to open it with
VMD from the directory in which you downloaded the animations; then, you
will load this animated_mode_007.vmd via "Load Visualization State" in File
menu of VMD, which will start playing the animation automatically using a
nicer graphical representation of your molecule.
*2) Efficiency of image analysis using normal modes:*
There is no GPU processing in HEMNMA. Regarding CPUs, it is not clear to me
whether you have a cluster with 24 processors Intel(R) Xeon(R) Silver
4214R, meaning that you can use up to 576 threads on this cluster (24 x 24,
as each of the Intel(R) Xeon(R) Silver 4214R processors can use up to 24
threads).
To get a quick view of the continuous heterogeneity in your dataset, you do
not need to process the entire stack of ~100k particles and you may resize
them by binning them by 2 or 4 as I will explain below (but not window
them, as this may cut out your particles). We could notice that a subset of
10k-20k particles is enough to examine the degree of the heterogeneity of a
larger dataset. How much to bin depends on how coarse and quickly you want
to explore the heterogeneity, but also on the amplitude of the
conformational change of your molecule. Let me explain this via these 2
points:
2.1 HEMNMA involves a 2-level multiresolution-pyramid image analysis. For
an input image of size N x N pixels, this means that HEMNMA first analyzes
the same image at a coarser resolution (N/2 x N/2 pixels) and then
propagates the result obtained quickly at this level onto the original-size
image level (N x N pixels) to refine the result. Depending on the degree of
the flexibility of the molecule, it may or may not be Ok to provide HEMNMA
with images of size 64 x 64 pixels (the conformational variability should
be detectable at the coarser level of 32 x 32 pixels that HEMNMA will
internally use). If conformational variability is likely to be detected at
the coarser level of 32 x 32 pixels used by HEMNMA internally (because the
amplitude of the conformational change is large), then you may input 64x64
pixel images to HEMNMA.
2.2 In HEMNMA graphical interface, you need to choose the rigid-body
alignment method ("projection matching" or "wavelets & splines") to use in
combination with the elastic alignment based on normal modes. The
"projection matching" method is faster but less robust to noise than the
"wavelets & splines" method. Additionally, it is useful to know that
"wavelets & splines" requires a power of 2 size of the image, meaning that
you may resize your image to 128 or 64 pixels but not to 100 pixels. With
"projection matching", there is no such restriction (you may input images
of size of 100 x 100 pixels, which may be interesting in your case).
It should be noted that the image resizing can be done directly with HEMNMA
graphical interface (step "Resize particles"). Note also that HEMNMA
currently allows entering individual particle images as well (beside a
stack form of input images). Although it may be interesting (or enough) to
first see the heterogeneity of a smaller subset of your full dataset (as I
said, 10-20k particles), you should also know that there are ways to split
your full (large) stack in several pieces and run these pieces on several
separate computers or run many parallel jobs on a supercomputer (each job
on a different piece of the stack) and then combine the results of the runs
on different pieces of the stack. The combined result file can later be
analyzed with HEMNMA graphical interface (the field "Precomputed results"
of the black/step "Image analysis with normal modes" that you can visualize
using "Advanced" button in "Expert level").
Last but not least, HEMNMA is part of ContinuousFlex plugin of Scipion (
https://github.com/scipion-em/scipion-em-continuousflex). We are regularly
updating this plugin and integrating new protocols, for SPA and ET. I would
like to invite you to update ContinuousFlex plugin in your Scipion
installation whenever the new updates become available (when an update is
available, red letters appear when you click on "scipion-em-continuousflex"
in Scipion's Plugin Manager).
I hope this helps. Let me know if you have any further questions.
Regards,
Slavica
On Wed, Nov 24, 2021 at 4:05 PM Carlos Oscar Sorzano <co...@cn...>
wrote:
> Dear Lizelle,
>
>
> I am forwarding your mail to Slavica, who is the person in charge of
> ContinuousFlex.
>
>
> Kind regards, Carlos Oscar
>
>
> El 24/11/2021 a las 15:58, Lizelle Lubbe via scipion-users escribió:
>
> Hi everyone,
>
> I would like to run HEMNMA on my particles and have done the NMA step
> using an atomic model. Is there any way to visualize the trajectories other
> than through VMD in Scipion? I am accessing the campus computer with
> Scipion remotely and it seems like VMD installation may be an issue.
>
> Three modes from an input PDB show high collectivity with low scores and I
> would like to now try NMA alignment with these modes. Given the high
> computational cost of the protocol, is it recommended to use a subset of
> particles for the analysis at present? The original stack is of ~100k
> particles (discrete heterogeneity was removed by 2 rounds of 2D
> classification and some 3D classifications but there is still continuous
> heterogeneity). It would be great to see the full range of variations with
> all particles but maybe not computationally feasible at present? The box
> size at 1.06A/pix is 256 so I can maybe bin by 4 to 64pix but windowing
> further will cut into my particle.
>
> We have 256GB RAM with 24 Intel(R) Xeon(R) Silver 4214R CPU @ 2.40GHz and
> 4 Quadro RTX 5000 GPUs.
>
> Kind regards,
> Lizelle
> Disclaimer - University of Cape Town This email is subject to UCT policies
> and email disclaimer published on our website at
> http://www.uct.ac.za/main/email-disclaimer or obtainable from +27 21 650
> 9111. If this email is not related to the business of UCT, it is sent by
> the sender in an individual capacity. Please report security incidents or
> abuse via https://csirt.uct.ac.za/page/report-an-incident.php.
>
> _______________________________________________
> scipion-users mailing lis...@li...https://lists.sourceforge.net/lists/listinfo/scipion-users
>
>
|
