Dear Jesus,

Be careful that the number of particles jumping between classes is very much dependent on orientation sampling. If your sampling is too coarse, particles will not be able to find their correct assignement. If you make your sampling finer and finer, you will see less and less particles jumping between classes. Just be careful that then, particles may be artificially stuck in a particular class. Sjors’s group suggests not going much finer than the orientation accuracy. An analysis of particle behavior should definitely be in that range to be meaningful.  

Hope this helps,

Amedee

--
Amedee des Georges, Ph.D. 
Assistant Professor, Structural Biology Initiative, CUNY Advanced Science Research Center
Assistant professor of Chemistry & Biochemistry, City College of New York
[log in to unmask] - http://asrc.cuny.edu  -  212-413-3232


On Jun 2, 2016, at 5:26 AM, Jesus GOMEZ <[log in to unmask]> wrote:

Thank you both for the fast responses,

Grigory, that is a really interesting paper. It does analyse the same phenomena I'm interested in, but their focus is in localizing particles that belong to the same class and have been separated, while my focus is to find particles that belong nowhere and have been aimlessly shifting classes. Anyway, their approach to merging classes seems quite useful. As you say, particles may jump between similar classes, and currently I try to take this into account by giving a intermediate score to those particles that travel frequently between the same few classes, but actually merging the classes might be a better approach. I believe I will either try to mimic their scrip in bash and add it to mine (with a proper authorship mention) or ask our IT team whether it would be possible to get matlab installed in my computer.

Reza, you are right about Sparx, is a quite versatile software. Sadly, our cluster seems to not like it so much, and we have some serious problems running some of the programs included, mainly when using MPI. I'm running a small test, but this far sxisac seems to run only using one cpu, so I expect it will take quite long to finish. Our IT team is working on this, but currently is not possible for me to use it efficiently. Still, knowing that this program does this is helpful, I can go into the files and related papers and try to find the criteria they use.

Thanks again for your helpful comments,

Best regards,

Jesus




On 2016-06-01 14:10, Reza Khayat wrote:
Hi,

ISAC does what you’re seeking and a lot more. It’s in the Sparx
package.

Best wishes,
Reza

Reza Khayat, PhD

Assistant Professor

Department of Chemistry

City College of New York

85 Saint Nicholas Terrace, CDI 2.318

New York, NY 10031

http://www.khayatlab.org/ [1]

212-650-6070

FROM: Collaborative Computational Project in Electron cryo-Microscopy
[mailto:[log in to unmask]] ON BEHALF OF Grigory Sharov
SENT: Wednesday, June 01, 2016 7:58 AM
TO: [log in to unmask]
SUBJECT: Re: [ccpem] Stability of particles in Relion classifications


Hi Jesus,

you might look into this [2] paper from Frank's group. The matlab
script described there should work (after few modifications) for the
latest Relion version.

Another thing is that particles will keep jumping if classes are very
similar. You could look into ccpem archives about discussion on 3D
classification convergence.

Best regards,
Grigory

--------------------------------------------------------------------------------


Grigory Sharov, Ph.D.

Institute of Genetics and Molecular and Cellular Biology
Integrated Structural Biology Department (CBI)
1, rue Laurent Fries
67404 Illkirch, France

tel. 03 69 48 51 00

e-mail: [log in to unmask] [3]

On Wed, Jun 1, 2016 at 1:34 PM, Jesus GOMEZ <[log in to unmask] [4]>
wrote:

Dear all,

Currently I'm working with Relion and I find myself in the
following situation: In each iteration there is a percentage (around
10-15% in the later iterations) of particles that changes classes.
What I don't know is if these particles are the same for all the
iterations or change every round of classification.

To address this my idea is to classify my particles based on their
stability along the 2D or 3D classification, meaning that those
particles that change classes few times will be considered good, and
those changing many times will be considered bad and removed from
the dataset. Note that this is different from the "particle sorting"
option in Relion, which only takes into account the correlation
between each particle and the class it belongs to for one given
iteration (my idea is to combine these two types of classification
for a more complete cleaning of the dataset). This is not very
difficult technically, since the data needed is in the *data.star
files outputted by Relion. In fact I have written a very simple bash
application that reads the data and awards points to the particles
based on events such as "staying in the same class", "times stayed
in the same class in a row" and such, placing more weight in the
later iterations, which are more relevant in my opinion.

My question is: does anyone use or know about something similar?
I'm looking for a solution for Relion, but any other program that
does this would also be welcome. Since I'm not a expert in this and
I'm using a quite simple and arbitrary approach to scoring the
particles I feel like there might already be a more established
solution for my problem that I have overlooked.

Thank you in advance,

Jesus Gomez de Segura



Links:
------
[1] http://www.khayatlab.org/
[2] http://embl.fr/%20http:/dx.doi.org/10.1016/j.jsb.2014.10.006
[3] mailto:[log in to unmask]
[4] mailto:[log in to unmask]