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