Tutorial - Tomographic Reconstructions in Priism (2/2)

This is part 2 of the tutorial for performing reconstructions of tomographic datasets with Priism; part 1 of the tutorial is at http://msg.ucsf.edu/em/EMNEW2/Tomog_tutor_v2.html.

5. Use EMTAR to Determine the Tilt Axis Orientation

If you already know this from other reconstructions at the same magnification or, for data collected with UCSF_TOMO, you believe the calibrated value, you can skip this step. The accuracy of this step is critically dependent upon the tilt offset being correct (the zero degree tilt is actually zero degrees). The mass-normalization step will usually do a good job of this, but you can do a better job when you view the backprojection later. Iterating through the whole process will therefore give you the best possible answer.

To determine the tilt axis orientation:

  1. In the EMTAR dialog, if you have not done so already, select the input tilt series with the "TiltSeries" button or the adjacent field. Also select the resolution of interest using the "Resolution" field (using the highest resolution possible (zero) might provide a better answer but will mean more computation time).
  2. Turn on the "Determine tilt axis" step in the EMTAR dialog.
  3. In the tilt axis determination parameters:
  4. Press the "Do It" button at the bottom of the EMTAR's dialog to generate the alignment parameters which include the calculated value for the tilt axis orientation. For the sample dataset, it estimates the tilt axis to be at an orientation of -79.94 degrees.

6. Use EMTAR to Compute the Alignment Shifts and Align the Tilt Series

We'll use three stages in this step: one to compute the shifts for the alignment using cross-correlations, another to apply the alignment parameters to generate an aligned tilt series, and a final step to reconstruct a central slice of the 3D volume. To set things up, use the following:

  1. If you adjusted the resolution level for the tilt axis determination calculation, set the value back to two in the main EMTAR dialog so the subsequent sections use the second lower resolution (500 x 500) in the input tilt series. Turn off the tilt axis determination step.
  2. In the EMTAR dialog, turn on the "Alignment" step. Stick with the default parameters for the first run: those defaults will use the tilt offset and tilt axis determined in the previous steps and will overwrite the input alignment parameters. When you have problems with the alignment, the option in the alignment step's special parameters to generate images of the cross-correlations can be useful. By changing the Wiener parameter for the phase weighting and the 2nd peak threshold level, you may be able to overcome some alignment problems. The stretching, filtering, and phase weighting options may also help in a few cases. Below are example cross-correlation results for the sample data set. The image on the right is the cross-correlation between the 7.99 degree and 9.99 degree projection (section 36 in the image file) and the image on the left is the cross-correlation between the 43.98 degree and 45.98 degree projection (section 54 in the image file). cross-correlation images
  3. Turn on the "Apply parameters" stage in the EMTAR dialog. That stage uses the alignment and mass-normalization parameters to generate a new tilt series where the projections have been interpolated to an aligned coordinate system and the intensities have been normalized. In the parameters for that stage, change the name of the output, in the "AlignedSeries" field, so the last part of the file name is "Cent_ufix_001_b4.MnAln" rather than "Cent_ufix_001.MnAln" to distinguish that the aligned series will be binned down by a factor of 4 compared to the input tilt series. Also, change the menu next to the "Image formation" button to read "scaled linear". That is an appropriate choice for the sample dataset. For other datasets, the logarithmic option may be a better choice. The linear option does not normalize the intensities at all. In the special parameters for the "Apply parameters" stage change the pcBase value to 0.001.
  4. Turn on the "EWBP" step in the EMTAR dialog. EWBP is an acronym for elliptically weighted backprojection. It is the less computationally intensive reconstruction available in the EMTAR interface; the other reconstruction method is TAPIR. In the EWBP parameters, change the name of the generated reconstruction, in the "Reconstruction" field, so the last part fo the file name is "Cent_ufix_001_b4.xzyw" rather than "Cent_ufix_001.xzyw". Adjust the "Y range reconstructed" to be "996 999". Because the second lower resolution (i.e. binned by a factor of 4) of a 2000 x 2000 tilt series is used, that choice of values will generate a single central slice through the reconstruction volume. Set z size of the reconstruction, which is the second value in the "Output XZ size" field", to be 600 rather than the default value of 150.
  5. Press the "Do It" button at the bottom of the EMTAR dialog to generate the alignment parameters, aligned tilt series, and reconstruction slice.

When the processing completes, the "Display" menu in EMTAR's main dialog (see Image 5 for a screenshot) is a convenient shortcut for viewing the results in a Priism window. With the exception of the "MASSNORM fit" option, the choices in that menu are shortcuts for bringing up CopyRegion for the file in question.

Image 6. Shows a screenshot of Priism and EMTAR highlighting EMTAR's "Display" menu. screenshot

After this stage of processing, you would usually look at the reconstructed slice and the aligned and normalized tilt series. I'll discuss the reconstructed slice in the next section. By playing animations of the aligned tilt series (the Movie option in the Image menu of the image window is convenient for that), you can qualitatively assess the alignment. Below is a QuickTime move of the aligned and normalized tilt series from the sample data.

Movie 2. Displays the aligned and normalized tilt series for the sample centrosome data. The QuickTime movie can be downloaded from http://msg.ucsf.edu/em/EMNEW2/movies/mnaln_v2.mov. It is 6.7 megabytes long.

7. Correct Problems with Centering and Angles

The image below is the reconstructed slice computed for the example dataset in the previous step. When looking at slices like this, focus on the following:

Image 7. Shows the central reconstructed slice for the example dataset. central slice

For the example image, the contrast is poor, but the features (for instance, the broad smudge about 30 to 70 pixels from the left edge and another, similarly shaped smudge another 50 pixels to the right of that) are roughly centered in the vertical bounds of the slice. It is difficult to see the bounds of the sample in the image, so I'd not make an adjustment to the tilt offset on the basis of this slice along. The missing wedge artifacts appear okay.

To apply corrections for the tilt offset or centering (for problems with the alignment, you'll want to check the tilt axis orientation value and, if you didn't run the tilt axis determination yet, run it to check against the value you had used) do the following:

  1. In the EMTAR dialog, turn off the alignment stage.
  2. Make sure the apply parameters stage is on in the EMTAR dialog. In the parameter dialog for that stage, add four times the shift in z you noted from the reconstruction slice to the third value in the "Output shift" field. The factor of four is there because the reconstruction was done at two resolution levels lower than the full resolution; in general, the multiplicative factor would be two raised to the selected resolution level. In the special parameters for that stage, enter the tilt offset you determined from the reconstruction slice.
  3. Make sure the EWBP stage is on in the EMTAR dialog.
  4. Press the "Do It" button at the bottom of the EMTAR dialog to regenerate the aligned and normalized tilt series and the reconstruction slice.

With the regenerated slice, look again to see if the sample features are centered vertically in the slice and that the sample appears to be approximately horizontal. If not, repeat the process as necessary until the sample does appear to be centered and horizontal in the reconstruction slice.

If you found that you had to adjust the tilt offset to get a horizontal sample in the reconstruction slice, take the tilt offset you determined and rerun the alignment and apply parameters steps:

  1. In the EMTAR dialog, turn on the alignment stage. In the special parameters for that stage set the tilt offset to the value you determined above.
  2. In the EMTAR dialog, make sure the apply parameters stage is on. In the special parameters for that stage, set the tilt offset to zero (when the alignment is run it will incorporate the offset into the tilt angle values in the alignment parameters; therefore, you don't want to add a further adjustment in the apply parameters stage).
  3. Turn off the EWBP stage unless you want to double check the appearance of the reconstruction slice.
  4. Press the "Do It" button to regenerate the alignment and aligned tilt series.

8. Use EWBP to Reconstruct the Full Volume

Up to now we've only reconstructed a single slice to avoid the overhead of computing the full reconstruction while adjusting some key parameters. We'll now proceed to reconstruct the full volume with EWBP:

  1. In the EMTAR dialog, turn off all stages except EWBP (we already have an up-to-date aligned and normalized tilt series from the previous step).
  2. In the EWBP parameters, set the "Y range reconstructed" to cover the full range of the input tilt series. For the example data which is 2000 x 2000, enter "0 1999" in the "Y range reconstructed" field.
  3. Press "Do It" at the bottom of the EMTAR dialog to generate the reconstruction.
  4. Since the reconstruction is in a flipped orientation, you typically also want the unflipped orientation. Use Priism's Flip (in the Priism menus, go to the "DataViews" menu, select "Rotation", and then select "Flip" to launch it) to generate the unflipped view.

The first movie below steps through the xz slices of the reconstruction for the example data. The second movie steps through the xy slices of the same reconstruction.

Movie 3. Displays the EWBP result (.xzyw) for the example dataset. The QuickTime movie can be downloaded from http://msg.ucsf.edu/em/EMNEW2/movies/xzyw_v2.mov. It is 9.7 megabytes long.

Movie 4. Displays the flipped version of the EWBP result for the example dataset. The QuickTime movie can be downloaded from http://msg.ucsf.edu/em/EMNEW2/movies/xyzw_v2.mov. It is 12.9 megabytes long.

9. Iterative Refinement of the Alignment and Reconstruction

Use this step to improve the alignment parameters and reconstruction that you obtained earlier. This step is significantly more computationally expensive then the earlier steps. I recommend running the process in parallel on a cluster or multiprocessor machine.

  1. In the Priism menus, select "Processing", and then "EM Processing", and finally "Iter. Alignment + Reconstruction" to open the dialog for the iterative refinement of the alignment and reconstruction.
  2. Select the same tilt series (using either the "TiltSeries" button or the adjacent field) that you used in the earlier steps.
  3. Use the "InAlignParam" or the adjacent field to select the file with the input alignment parameters. For the example dataset, they are in Cent_ufix_001.bprmMn in the same directory as the input tilt series.
  4. The output alignment parameters (you will get one set per every iteration of the algorithm) and the output reconstruction are, by default, written to the same directory as the tilt series. If you want to change that, edit the fields labeled "OutParamBase" and "Reconstruction". To flag that the reconstruction have been binned by a factor of four from the full resolution, change the final part of the name in the "Reconstruction" field from "Cent_ufix_001.xzy" to "Cent_ufix_001_b4.xzy".
  5. Use the same resolution level, 2, that you used earlier by entering 2 in the "Resolution" field.
  6. The menu next to the "Apply massnorm" button selects the image formation model to use Use the same model as you used with EMTAR (for the example dataset, that was the "scaled linear" model).
  7. In the "Reconstruction shift" field, enter the same values as you used for the full reconstruction in the previous step.
  8. In the "Reconstruction Z size" field, you can use the same size as you used for the full reconstruction in the previous step. For cases where you could see the sample thickness in the earlier reconstruction, you could use that (multiplied by by the resolution factor, two raised to the selected resolution level) with some added slop (perhaps 20% of the total though it hasn't been rigorously tested to see what is optimal) as the size.
  9. Enter ten for the number of runs.
  10. In the special parameters, change the setting for "pcBase" to be .001. The remaining special parameters are likely fine.
  11. Press the "Do It" button at the bottom of the main dialog to start the processing.

At the end of the processing, you'll have a series of ten alignment parameter files (one for each iteration) and a reconstruction.

10. Stepping up to Higher Resolutions

In the steps so far, you used a version of the tilt series binned down by a factor of four. For a quick reconstruction (no further refinement of the alignment parameters) using a higher resolution version of the tilt series to get a higher resolution reconstruction you would:

  1. Open the EMTAR dialog.
  2. Select the tilt series.
  3. Enter the resolution level you want in the "Resolution" field.
  4. Turn on the apply parameters processing stage. Use the best alignment and mass-normalization parameters that you determined at the lower resolution. Use the same values for the image formation model and, in the special parameters, pcBase that you used for the lower resolution data set.
  5. Turn on either the EWBP reconstruction method or the TAPIR method (takes more time but applies a positivity constraint which should improve the resolution). Enter the same reconstruction shift and size that you used for the lower resolution reconstruction in the "Size (XYZ)" and "Shift (XYZ)" fields in EMTAR's main dialog.
  6. Press "Do It" at the bottom of EMTAR's dialog to generate the higher resolution aligned tilt series and reconstruction.

To get more out from stepping up to a higher resolution, you will need to refine the alignment parameters at that resolution. The procedure is essentially the same as the iterative refinement at the lower resolution except:

Part 1 of the tutorial