refinement questions

Submitted by Yanyan Zhao on Tue, 2015-06-09 14:23

Forums

Frealign

I have a problem extracting a class of particles after multi-refinement using Frealign. The multi-refinement gave three classes with about 15,000 particles in each class. But when I use merge_classes.exe to extract the thirst class of particles, I only got around 8,000 particles. Why is that?

This is the result when I run frealign_calc_stats.
1: N = 14852 Ave. OCC = 33.1751 Ave. LL = -17671.9 Ave. Score = 32.5013
2: N = 14293 Ave. OCC = 31.9257 Ave. LL = -17812.8 Ave. Score = 31.8804
3: N = 15624 Ave. OCC = 34.8992 Ave. LL = -17792 Ave. Score = 32.3776
44770 particles, average LL = -17758.8, average Score = 32.2598

the input when I run merge_classes.exe are:
input stack: the original image stack for refinement which has 44,000 particles in it.
the number of classes merged: 1
input parameter file: .par file in 18th refinement cycle
OCC: 100.0
min score: 0.00

the mparameter file is like this:

# Refinement-specific parameters
MODE 1
start_process 2
end_process 18
res_high_refinement 15.0
res_high_class 8.0
thresh_reconst 0.0
nclasses 3

data_input frealig
raw_images frealig_stack
image_contrast N
outer_radius 150.0
inner_radius 0.0
mol_mass 800.0
Symmetry D8
pix_size 1.071
dstep 6.0
Aberration 4.7
Voltage 300.0
Amp_contrast 0.1

# Expert parameters (for expert users)
XSTD 0.0
PBC 50.0
parameter_mask "1 1 1 1 1"
refineangleinc 4
refineshiftinc 4
res_reconstruction 0.0
res_low_refinement 0.0
nbootstrap 1000
FMAG F
FDEF F
FASTIG F
FPART F
FFILT T
FMATCH F
FBEAUT F
FBOOST F
RBfactor 0.0
beam_tilt_x 0.0
beam_tilt_y 0.0
mp_cpus 1
restart_after_crash F
delete_scratch T
qsub_string_ref ""
qsub_string_rec ""
first_particle
last_particle
frealign_bin_dir
scratch_dir

Alexis Tue, 2015-06-09 14:42

Hi Yanyan,

In merge_classes.exe, the question about occupancy asks for the threshold occupancy. In other words, particles with occupancy lower than 100% (this is what your script uses) will not be included. When you run frealign_calc_stats, particles are "assigned" to a class if their occupancy for that class is higher than for other classes, but they may have occupancies lower than 100%. I think this may be where the discrepency arises- there are probably ~ 7,000 particles which have very high occupancies, but not quite 100%.

In any case, if you're trying to "extract" a class of particles, perhaps you should use the select_classes program instead of merge_classes?

Alexis

Yanyan Zhao Tue, 2015-06-09 15:03

Thank you, Alexis. How do you deicide the threshold occupancy in this situation?

Alexis Tue, 2015-06-09 15:16

That would depend on what you're trying to achieve.
If you are trying to obtain a new image stack & parameter file with only particles from one of the classes, I would use select_classes.exe, rather than merge_classes. In that case you won't have to select an occupancy threshold.

If you really are merging classes or you want to use merge_classes for some other reason, maybe you could go to 90%. In most cases, that'll catch most particles.

Yanyan Zhao Tue, 2015-06-09 15:29

So if I useselect_classes.exe, the input file are:
input stack: stack with all of the particles
number of classes to select from: 1
input parameter file: the parameter file for one class

Am I doing this correctly?

Alexis Tue, 2015-06-09 15:53

Just run it interactively and you'll see what the questions are...

It'll ask you how many classes there are, then for each class, it'll ask you

  • the parameter filename
  • whether to select this class
  • if you select this class, the output parameter filename

Then it'll ask you

  • the output stack filename
  • whether to resest the occupancies in the output par file to 100.0