[3dem] Particle picking within Relion 1.3 using projection images

[mf2944 at columbia.edu]

Dear all,

I recently tried using the particle picking within Relion and if I use  
the recommended procedure doing first a small 2D classification,  
followed by autopick, the program works fine.
However, if I make a 2D classification from a few micrographs, the 2D  
classes are only representing some major orientations that I have in  
my sample. But for the more rare orientations (those I would  
especially like to pick more efficiently) I do not obtain a 2D class.  
These only come up if I make a complete 2D classification using all  
micrographs.
Therefore I though to use 2D projections from an initial lower  
resolution volume I obtained using Relion.

To create these projections i used xmipp
xmipp_image_convert -i xxx.mrc -o xxx.vol -t vol
xmipp_angular_project_library -i xxx.vol -o projection.mrc --sym x  
--sampling_rate 22.5

Since the pixel size of the input is twice as big as the pixel size of  
the new dataset I am working is rescaled the projections
xmipp_image_resize -i projection.mrc -o projection_scaled.mrc --factor 2

However, using these as input in the relion particle picking is not  
successful.
My feeling is that the FOM for the true particles in the picking are very low.
As additional information I calculated the statistics of the reference  
images create in 2D classification vs xmipp projection

projections
Total number of images/volumes: 13
min= -0.912874 max=  2.861991 avg=  0.022331 stddev=  0.633817

2D classes
Total number of images/volumes: 10
min= -0.481988 max=  0.763749 avg=  0.055558 stddev=  0.249007

Could anybody help me out with what I could do to get this working? Or  
am I missing something essential that this route will never work?

Best regards
Marcus


-- 
Marcus Fislage, PhD

Howard Hughes Medical Institute (HHMI)
Columbia University
Department of Biochemistry and Biophysics
Lab of Joachim Frank
New York, NY

Phone: 212.305.9524
Fax: 212.305.9500