Three Dimensional Reconstruction
of Single Particle Specimens using
Reference Projections
This page describes a procedure for creating a 3D ribosome structure.
An automated particle selection process serves to identify ribosomes
from digitized electron micrographs. The image series is then aligned
relative to reference projections through shifts and rotations using
either AP MQ or AP MR command. A 3D iterative reconstruction is calculated. Difference map and its
significance can be calculated. Some other useful
methods are described as well.
updated 10/27/97, Yiwei Chen
updated 5/20/97, Amy Heagle
updated 2/5/97, Paul
Automated Particle Selection
When you use the batch file in this document to do 3D reconstruction, first click the
file name and copy this batch file to your batch file. In the begining of each batch file
you will see which parameters should be changed and what the name of input, output files
should be. Some of the batch files will call corresponding procedures (just listed below
the batch files). You need not change anything in the procedures.After you create you batch
file, use following fomat to run the batch files:
spider the extension of batch file/the extension of micrograh file
the name of batch file,
for example
spider win/50s b01
to run the batch file of b01.win.
1.Adjust the micrograph dimensions - this step is only needed
for computers without SGI libfft library.
Use the Spider command FI to find the dimensions of the
micrograph. Check that the image dimensions can be interpolated
down to exactly 1/4 the original size and then verify that the
interpolated image size can be Fourier transformed (see list of
appropriate image sizes in the FT page of the index of spider
operations). If the original image dimensions need to be
changed, window the micrographs accordingly.
b01.win
1a.If you need to know the CTF estimate the power spectrum:
b01.pws
power.pws
By the above batch file, you can get the 1D profile of power spectrum
named "pw_ro***". Using TF DDF command to determine DeFocus and
amplitude contrast. Whe you choose only one point of minimum, you will
input amplitude contrast (usually 0.09 or 0.1). Otherwise, the amplitude
contrast will be calculated.
2.Select a background noise file
Open one micrograph in WEB and use the Pixel utility to identify
coordinates for a window containing background noise
(no particles). Window this region from the micrograph.
b02.noi
3.Create a mask file
Create a circle with dimension and radius (generous) corresponding
to the particle size.
b03.mod
4.Run automatic particle selection
The micrographs are windowed to the size divisible by 4,
interpolated down by 1/4, Fourier filtered
with a Gauss-low pass, peak searched over regions
corresponding to particle dimensions, particles are windowed
out, centered, peak locations are corrected and partciles
are windowed again.
b04.rmp and
mpk.rmp
5.Verify automatically selected particles
Sort through the output files to eliminate any non-particles.
This is accomplished in WEB with the Categorize command by
montaging the particle image files, manually clicking on each good
particle, and then saving the good file numbers in a doc file.
The doc file then needs to be adjusted so that the image numbers
are in ascending order with sequential key numbers using the
following program.
b05.ati
At this point either Multireference Alignment using AP MQ
(preferably) or Multireference Alignment using AP MR can be
used. Follow either track to Iterative 3D reconstruction.
Multireference Alignment using AP MQ
6.Create a selection doc file for 83 reference projections
Reference projections are views of ribosomes collected in
a previous project. This file contains a column of numbers
ranging from 1-83 which will be used to call the reference
projections used in the alignment of the particles in step 9.
b06.spl
7.Obtain reference projections from a reference volume
b07.pjq
Angles corresponding to reference projections are located in
angf
and were generated using delta(theta)=15.
8.If the doc file containing the good particles is large,
it can be broken up into manageable parts so that step 9 can
be run on many machines.
b08.ord
9.Align the particles to the reference projections AP MQ
This is a multireference alignment of an image series.
b09.amq
10.Combine all resulting alignment files into one doc file
b10.dqi
11.Rotate particles according to alignment parameters
Any particles that correspond to reference projections 84-164
are also mirrored since projections 84-164 are mirror images
of the first 83.
b11.rpa and
alq.rpa
12.Compare aligned particles to reference projections
Create a doc file which identifies particles into reference
projection groups and displays the correlation coefficient
describing the relative similiarity of the particle to the
reference projection.
Warning: correlation coefficient is not normalized.
b12.clq
Note: Instead, faster Fortran program can be used:
group.f
/net/penang/usr1/pawel/useful-fortran-programs/group.exe
The particles corresponding to each projection can be viewed in
WEB under the Montage from doc file selection to visually
associate a particular correlation coefficient to the image.
The greater the value, the more similiar the particle to its
reference projection. Identify a minimun correlation coefficient
that describes true particles as opposed to erroneously selected
particles.
13.Again compare particles to reference projections, this time using
particles above a specified correlation coefficient. Repeat
b12.cla, eliminating particles with low correlation coefficients.
Again, montage from each reference projection doc file the aligned
particle images. Select Compute Average to view an average of
the particles displayed.
14.Compute averages for all projection groups
It is sometimes convenient to view all averages together by
montaging them in WEB.
b13.avg and
avg.rav
15.In case of strong overrepresentation of some of the angular
directions the numbers of images per directions can be limited
to certain number
b14.eqp
Or keep half of the best per direction, but no more
than specified number.
b14.eqs
To further investigate particle classification go to
Other Methods
Go to 3D reconstruction
Multireference Alignment using AP MR
6.Create a selection doc file for 156 reference projections
Reference projections are views of ribosomes collected in
a previous project. This file contains a column of numbers
ranging from 1-156 which will be used to call the reference
projections used in the alignment of the particles in step 9.
b06.sel
7.Obtain reference projections from a reference volume
b07.pjq
Angles corresponding to reference projections are located in
angr
and were generated using delta(theta)=15.
8.If the doc file containing the good particles is large,
it can be broken up into manageable parts so that step 9 can
be run on many machines.
b08.ord
9.Align the particles to the reference projections AP MR
This is a multireference alignment of an image series through
shifts and rotations.
b09.apr
10.Combine all resulting alignment files into one doc file
b10.dli
11.Rotate particles according to alignment parameters
Any particles that correspond to reference projections 85-166
are also mirrored since projections 85-166 are mirror images
of the first 84.
b11.alm
12.Compare aligned particles to reference projections
Create a doc file which identifies particles into reference
projection groups and displays the correlation coefficient
describing the relative similiarity of the particle to the
reference projection.
b12.cla
Note: Instead, faster Fortran program can be used:
groupa.f
/net/penang/usr1/pawel/useful-fortran-programs/groupa.exe
The particles corresponding to each projection can be viewed in
WEB under the Montage from doc file selection to visually
associate a particular correlation coefficient to the image.
The greater the value, the more similiar the particle to its
reference projection. Identify a minimun correlation coefficient
that describes true particles as opposed to erroneously selected
particles.
13.Again compare particles to reference projections, this time using
particles above a specified correlation coefficient. Repeat
b12.cla, eliminating particles with low correlation coefficients.
Again, montage from each reference projection doc file the aligned
particle images. Select Compute Average to view an average of
the particles displayed.
14.Compute averages for all reference projections
It is sometimes convenient to view all averages together by
montaging them in WEB.
b13.avg
15.In case of strong overrepresentation of some of the angular
directions the numbers of images per directions can be limited
to certain number
b14.eqp
Or keep half of the best per direction, but no more
than specified number.
b14.eqs
To further investigate particle classification go to
Other Methods
Go to 3D reconstruction
Iterative 3D Reconstruction
16.After deciding on a correlation coefficient threshold, create a
selection doc file which refers to the particles to be used in
the 3D reconstruction.
b15.par
16a. (APMQ) b15.rar and
par.rar
16b.If step 15 was necessary, create a new doc file of particles.
b15.new
17.Create a doc file containing particle file numbers to be used in the
3D as well as reference angles for each particle
b16.ang
17a. (APMQ) b16.ran and
anq.ran
18.Split select file used in 3D into two separate select files to be
used in the following two 3D reconstructions for comparative
purposes.
b17.rod and
ode.rod
19.Compute the 3D reconstruction of half of the available particles.
b18.bpe
20.Compute the 3D reconstruction of the other half of the available
particles.
b19.bpo
21.Compare the two half volumes
b20.rff
22.In UNIX, use gnuplot to view the resolution curve
plot 'doccomp.ext' using 3:5 with lines
an online manual for gnuplot:
http://www.cs.dartmouth.edu/gnuplot_info.html
23.Compute the 3D reconstruction
b21.bpr
24.Using different correlation coefficients, create various volumes by
repeating steps 16 through 23.
25.3D projection alignment
Compute a projection of the final volume, calculate distances
between projections, and convert output to angular doc file.
Calculate new, refined 3D structure using centered projections
and the corrected angles from the angular doc file.
The performance of the procedure (and the quality of the
reconstruction) can be judged by the average correlation
coefficient between projected structure and the input data.
It is stored in the last line (key -1) of the shift files.
b22.prj
uses procedures:
Makesel.prj
align3d.prj
Makeselsort.prj
Old version, do not use: b99.prj
25a.Evaluate the angular distribution of projections
b98.dis
25c.3D projection alignment with the CTF correction
If more than one defocus group is available and
CTF was estimated (defocus values are known) run 3D projection
alignment with the CTF correction:
b52.prj
uses procedures:
Makesel.prj
align3d.prj
Difference Maps
26.Create difference map.
b23.dif
Significance
Even though the variance in 3D cannot be calculated it is possible
under certain simplifying assumptions to assess significance of the difference
map in 3D. It has to be assumed that the structure has only one region
that varies (this region can be composed of many subregions) and that
there is no correlation between changes of this region and any other
changes in the structure. Under these assumption a set of 2D t-tests
(say k, as many as there were 2D reference images used) can be performed
on between 2D projections of control and investigated structures. Thus, aach
3D voxel can be related to k results of such tests. Using Bonferroni's
adjustment the overall difference in 3D is significant at the level
a if the sum of k t-test results is a. In other
words all the individual t-tests in 2D should be significant at the
level a/k.
27.Average the aligned particles and compute variance.
b24.avg
28.Two data set are likely to be normalized in a different
way, thus it is better to subtract average value from
all the averages.
b25.nrm
29.Compute t-tests for all the pairs.
b26.ttt
30.Backproject 2D significance projections into 3D volume.
b27.bpd
31.Threshold and mask volume. Thresholding should be done according
to the prescription above.
b28.thm
Other Useful Methods
Classification of Particles
This method can be used to investigate differences between particles.
Aligned particles are grouped according to associated view and then
subdivided into classes within each view based on similiarity.
Copy each particle into its associated view subdirectory (1-87).
b50.grp
Run correspondence analysis for each of the 87 views.
b51.csi
Run second part of correspondence analysis.
b52.cas
Classification
b53.clk