2d classification - blurry and seems to become worse after 10 iterations

[WVP]

Dear Steve,

I am struggling with a dataset of a small protein (144kDa) and I don't know which step would be best to do now. I used Eman for boxing and used Relion for 2dclassification. The first averages, didn't look bad, but the program seemed to be unable to throw bad particles away to bad classes even after variating box (80) and masking size (60). I therefore opted for a better selection of the micrographs and particles. I used the Wiener filtered particles to discriminate good from bad particles and deselected all the crowded particles. This new selection made it also clear that the classes seen make sense as the particles match with these averages. The new 2d classification however doesn't really improve- they stay blurry and after iteration 15 the alignment seems to become more noisy. I have attached some images as attachment (relion it 10 and 25 and e2refine2d).

best wishes and apologies for all the questions,

Wouter

[Steven Ludtke]

Hi Wouter. To clarify, were the EMAN class-averages you show done after or before you tried to eliminate the 'crowded' particles ?  The EMAN class averages (which are done without a mask, unless you impose it yourself) seem to imply a significant amount of particle crowding still occurring. While it is possible to process data under somewhat crowded conditions through use of aggressive masking, and insuring that your particle picking is well-centered, it isn't optimal. Crowding can cause a number of problems in EMAN, particularly with SSNR estimation and centering of particles. Using other software doesn't really solve this problem, the problem will simply impact you in different ways. The best advice if you have heavy crowding is to return to the bench and target grids with (much) lower particle density. That is not to say it is completely impossible to process the existing data or get any results at all, just that the amount of effort you spend struggling to get the image processing to work around the problem in many cases is more than it would take to improve your grids, and I can virtually guarantee that you could achieve more robust structures at better resolution if you improve the data quality in this way.

That said, if you really need to process the existing data, there are a few tricks you can use:

1) Insure that your boxed out particles are well-centered. ie - one of the several particles which may be present in the box should be within a few pixels of the center of the box

2) use the file browser and select a particle stack from one of your images, and hit the 'filtertool' button 

(There is a video tutorial on this here, if you aren't familiar with FilterTool : https://youtu.be/mTJ4gH7EOUA?t=26m1s)

3) In FilterTool, add a "filter.lowpass.gauss" processor, followed by a "mask.soft" processor

- The filter.lowpass.gauss will allow you to make the particles blurrier so you can see them more easily. Adjust this to a convenient level for visualization

- You can then interactively adjust outer_radius and width on the mask.soft processor to find parameters which mask out the particles effectively in all orientations without having a sharp edge, or impeding on the particles themselves. 

- once you have mask parameters you are satisfied with, make a note of the outer_radius and width, then close filtertool

4) In CTF->Generate Output, there are 3 empty text boxes. 

- in the first of these, put "normalize.circlemean:radius=XX" where XX is the outer_radius you determined above

- In the second of these, put:  mask.soft:outer_radius=XX:width=YY  (replace XX and YY with your values). 

- Make sure phase flipping is selected, and that your micrographs are selected just as they were when you last ran this program, then launch the task.

5) For each particle stack, this will produce an __proc variant of the file which has now has a soft mask applied to it.

6) Any particle sets you generated previously will lack the __proc variant, so rerun this step as well

7) now try running e2refine2d on the __proc variant of your particle set. This should give you somewhat nicer looking averages. Don't expect them to appear to be 'high resolution' but they should at least focus more on the different orientations of your particle than on the different variations of crowding present in your data

8) make an initial model from these new class-averages, and try an initial 3-D refinement, again using the masked particles.

If anything goes wrong in this process, please just ask about it

--
--
----------------------------------------------------------------------------------------------
You received this message because you are subscribed to the Google
Groups "EMAN2" group.
To post to this group, send email to em...@googlegroups.com
To unsubscribe from this group, send email to eman2+un...@googlegroups.com
For more options, visit this group at
http://groups.google.com/group/eman2

---
You received this message because you are subscribed to the Google Groups "EMAN2" group.
To unsubscribe from this group and stop receiving emails from it, send an email to eman2+un...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
<firstaverages.pdf>

----------------------------------------------------------------------------
Steven Ludtke, Ph.D.
Professor, Dept. of Biochemistry and Mol. Biol.                Those who do

Co-Director National Center For Macromolecular Imaging            ARE

Baylor College of Medicine                                     The converse
slu...@bcm.edu  -or-  ste...@alumni.caltech.edu               also applies
http://ncmi.bcm.edu/~stevel

[WVP]

Dear Steve,

thanks for the information. The classes were already after deselecting the particles which were to crowded - I removed the once where the particles were really touching the other particles. I have another microscope session in August (we depend on the generosity of other labs to have access to a good microscope), so meanwhile I'll try the things you've suggested.

best wishes and many thanks,

Wouter

Op dinsdag 16 juni 2015 15:50:23 UTC+2 schreef Steve Ludtke: