Particle Picking
1. Selecting a box size
All of the projection-matching and reconstruction involves multiple
Fourier transforms per particle. It is therefore a good idea to find
a box size that transforms fast (Note that simply expanding the image
size to a power-of-two size does not necessarily mean the fastest transformation).
The program bfft has an option to test the speed of Fourier
transformation for different image sizes:
bfft -test 2,250,260
Timing the execution of fast Fourier transforms: FFTW planner option: 0 Number of dimensions: 2 Size range: 250 - 260 Size Ptime Etime Prime factors 250 0.00723 0.08392 2 5 5 5 251 0.03657 0.08565 251 252 0.01181 0.11592 2 2 3 3 7 253 0.07655 0.00492 11 23 254 0.06900 0.23176 2 127 255 0.15219 0.00303 3 5 17 256 0.00438 0.04742 2 2 2 2 2 2 2 2 257 0.10456 0.08203 257 258 0.22105 0.09444 2 3 43 259 0.09175 0.00598 7 37 260 0.00661 0.08395 2 2 5 13 Ordered times: Size Etime 255 0.00303 253 0.00492 259 0.00598 256 0.04742 257 0.08203 250 0.08392 260 0.08395 251 0.08565 258 0.09444 252 0.11592 254 0.23176
The FFTW3 library used for Fourier transformation has an optimization capability that is typically included when it plans for a particular size transform. This can be switched on for the tests:
bfft -test 2,250,260
Timing the execution of fast Fourier transforms: FFTW planner option: 1 Number of dimensions: 2 Size range: 250 - 260 Size Ptime Etime Prime factors 250 0.50410 0.00204 2 5 5 5 251 0.86033 0.00336 251 252 0.57269 0.00184 2 2 3 3 7 253 1.45545 0.00182 11 23 254 1.59815 0.00177 2 127 255 2.72546 0.00144 3 5 17 256 0.29502 0.00122 2 2 2 2 2 2 2 2 257 2.36167 0.00198 257 258 3.26567 0.00167 2 3 43 259 1.30502 0.00184 7 37 260 0.59385 0.00128 2 2 5 13 Ordered times: Size Etime 256 0.00122 260 0.00128 255 0.00144 258 0.00167 254 0.00177 253 0.00182 252 0.00184 259 0.00184 257 0.00198 250 0.00204 251 0.00336
Now the planning time (Ptime) is much longer, but the execution time (Etime) is much shorter. The differences in execution time between the different sizes are also smaller, but still significant to do this test to pick a box size.
2. Picking particles in bshow
The fastest way to pick particles is to open a parameter (STAR) file for all the micrographs in bshow. To activate the creation of boxes for picking, click on the boxing tool in the tool panel (menu item "Window/Tools"). This will bring up a dialog box where the particle box size and bad area radius can be set. A typical particle can be selected and the box size set appropriately (the bad area radius is usually about a quarter or third of the box size). All the particles can be picked (left button) and deleted (shift-left button), and bad areas selected (control-left button or right button). The particle picking dialog box has buttons to move forward or backward through the series of micrographs referenced in the parameter file.
In the case of more than one micrograph per field-of-view, the particles in only one should be picked. This is typically the further-from-focus micrograph (usually the second one in each focal pair).
3. Finding particle locations in other micrographs in a field-of-view
Once all the particles have been picked in one micrograph in a field-of-view, the locations in the other micrographs can be found by aligning the micrographs to first one. This is most commonly done where the particles were picked in the further-from-focus micrograph and their locations need to be transferred to the closer-to-focus micrograph. All of the micrographs can be aligned using one command line:
bmgalign -v 7 -align mic -ref 2 -resol 2000,30 -correlate 1024,1024,1 -out klh_aln.star klh_*.star > klh_aln.log &
To check the alignment, look for the key word "Best fit" in the log file:
grep Best klh_aln.log
The last number on each line gives the RMSD in pixels of the alignment, where values below 5 pixels represent good fits, below 10 is still acceptable, but any larger values indicate incorrect alignment. Visual inspection of the particle coordinates in the closer-to-focus micrographs should be used to assess the success of alignment.
This alignment is not always successful, and the parameters that can be adjusted to improve it are the resolution limits (option -resolution) and the size of the tiles used for correlation (option -correlate, typically adjust it to larger tile sizes).
4. Particle extraction
Once all the particle coordinates are specified in the parameter files, the particle images can be extracted (change to the part directory first):
bpick -v 7 -extract 100 -back -norm -partpath ../part -partext pif -out klh_pick.star ../mg/klh_aln.star
The paths in the parameter file are very important for the programs to find the images. Several of the programs processing parameter files have -*path options that should be used to set the paths for micrographs, power spectrum images and particle images. If the images are written in undesirable directories, this should corrected with the various -*path options.
5. CTF correction
If desired, the extracted particle images can then be corrected for the CTF (change to the ctf directory first):
bctf -v 3 -data float -back -action flip -resol 2000,10 -partpath ../ctf -out klh_ctf.star ../part/klh_pick.star
All the modified particle image files should be written to the ctf directory.