Reconstructing a Cryo-Sample with Patch Tracking

(IMOD 5.1)

University of Colorado, Boulder

This tutorial will guide you through processing a cryo tilt series without fiducials, using the patch tracking method of alignment.  The tilt series is from a vitrified section and was acquired by Cedric Bouchet-Marquis.  It is recommended that you do the basic cryo tutorial or the high-resolution cryo tutorial before this one.

Getting started:

• Download the sample data set from our web site.
• Move the data set file "cryosectionData.tar.bz" to the directory where you want to work on it.  Its contents will unpack into a subdirectory named "cryosection".
• cd to the directory with the file
• Enter the command:
      imoduntar cryosectionData.tar.bz
  or, anywhere except on Windows without Cygwin, you can use
      tar -xjf cryosectionData.tar.bz
• Enter the data set directory with:
      cd cryosection
• Start Etomo by entering
      etomo
  and press Build Tomogram.

Tomogram setup:

• Press the file chooser icon on the Datatset name line and select the stack file.
• For SystemTemplate, select cryoSample.adoc.
• Press Scan Header
• Enter 0 for Fiducial diameter. Some parameters for X-ray removal are set based on the fiducial size to avoid erasing fiducials, so it is important to enter 0 rather than an arbitrary size when there are no fiducials.
• Select Parallel Processing to use multiple CPUs to compute the reconstruction, or select Graphics card processing to use the GPU of an Nvidia card.
• In the Axis A group box, select Series was bidirectional from and set the angle to 0.
• Press Create Com Scripts

Pre-processing:

• Press Create Fixed Stack.
• Press Show Min/Max for Fixed Stack.  There are no extreme values any more.
• If you still see some extreme values in a data set, you should lower the two criterion entries a little and run again. Ccderaser has already iterated the process three times, so iterating with the current criteria (by pressing Use Fixed Stack then Create Fixed Stack) will not remove more extreme values.
• Press Use Fixed Stack.

Coarse Alignment

• Press Calculate Cross-Correlation
• When done, press Generate Coarse Aligned Stack

Fiducial Model Generation:

In this step, patches of image will be tracked from one view to the next, and the center positions of the patches will be placed into a model file. This model can be used to solve for the tilt series alignment, much as a model of true fiducial markers can.

• Select Use patch tracking to make fiducial model. 
• Change the Size of patches(X, Y) to 680,680 (if it is not already set to that); this will give 4x4 patches.
• Press the Advanced button at the bottom and change the High frequency cutoff radius to 0.125 (if it is not already set to that).  The parameters for patch tracking are mostly based on the ones for coarse alignment since the same program is being run.  However, the correlations for coarse alignment are done on almost the full image and are generally binned down, so the high frequency filter cutoff of 0.25 is adequate for filtering out noise.  The patches being correlated are smaller and will not be binned, so stronger filtering is needed.  Press Basic to make it easier to see the basic controls.
• Press Track Patches.
• When done, press Open Tracked Patches to see the tracked locations on the image and in Model View.  No specific feature has been tracked, just the whole patch area.  Notice that there are just 16 contours going all the way through the tilt series.

Fine Alignment:

For this data set, the fine alignment process will involve cycling back to the patch tracking page.

• Press Compute Alignment.  The mean residual is fairly high, 1.22 nm.  Such a high residual occurs because some of the positions tracked through the patches do not correspond to projections from single points in 3D.  A better fit to the points can be gotten by breaking the contours into pieces, but first, see if better tracking might be achieved by setting an appropriate tilt angle offset.
• Right-click over the Fine Alignment panel and select Align log file in the popup menu.  Select the Surface Angles tab.
• The "Total tilt angle change"  implied by fitting a plane to the 3D locations of the patches is large (-12 degrees).
• Press View 3D Model and press the "F" hot key in the 3dmodv window to see the 3D patch positions in X/Z view.  The patches do fall in a plane at a ~12 degree angle, so the solved tilt angle change is reasonable.  When this angle is bigger than a few degrees, the patch correlations at high tilt can become significantly more accurate with the angle taken into account.
• Enter the angle in the Total tilt angle offset field; this will make the specimen nearly level in the tomogram in the positioning step.
• Return to the Fiducial Model Generation page, press the Advanced button and enter the same angle in the Tilt angle offset field.
• Press Track Patches then return to the Fine Alignment page and press Compute Alignment.  The mean residual has dropped to 1.01 mn, indicating that tracking was more accurate.
• Return to the Fiducial Model Generation page and select Break contours into pieces.
• Press Recut or Restore Contours.
• Return to the Fine Alignment page and press Compute Alignment.  The mean residual and leave-out error are much lower, partly because different 3D positions are now being solved for somewhat different points in the specimen, and partly because simply cutting up a contour ithat does represent a single point in 3D in this way allows smaller errors to build up over the shorter tilt ranges in each contour, reducing the errors.  Section 7.12 of the Tomography Guide presents estimates that the latter effect typically reduces the leave-out error by a factor of 0.7 (range 0.54-0.94).  Here, chopping up contours reduced the leave-out error by a factor of 0.32, so much of this reduction does represent an improvement in alignment.
• Press View/Edit Fiducial Model.  In the Model View window, rotate the model so that you can see all the tracks well by pressing the middle mouse button and moving the mouse at a steep diagonal.  Each track consists of 7 overlapping segments, which you can see by right-clicking near a track to select a contour; only a small segment is highlighted.
• Bring up the align log file and switch to the Coordinates tab.  The "mean resid" column shows the mean residual in each contour or segment of the track.  The range is large, from 0.06 to 1.14 nm.  When patch tracking was first developed, the approach was simply to delete the contours with the highest residuals, and the Bead Fixer has a Look at contours mode to assist this process. Not only is this somewhat arbitrary, but it would lead to trouble in this case, since 10 of the highest 11 residuals occur for contours passing through zero degrees.  The robust fitting that is now available is a better approach.
• Turn on Do robust fitting with tuning factor.  Etomo recognizes that this is a patch tracking model and has turned on Find weights for contours, not points.  Press Compute Alignment.  In the project log, you can see that the benefit from robust fitting is small, but at least it is positive.
• Open the align log again.  Each contour has been given a weight between 0 and 1, where a weight of 0 means that the points in the contour are ignored in the fitting.  On the Robust tab, a summary line indicates 9 contours have weights under 0.5.  On the Coordinates tab, there is a new column with the weights.  The contours that have small weights are ones that have large residuals relative to other comparable contours (i.e., ones at similar tilt angles).
• For your own data sets with patch tracking, you will probably want to try the robust fitting whenever you break contours into pieces.  If contours are not broken into pieces, there are usually too few contours to allow robust fitting.
• Press Run Cross-Validation to optimize the selection of variables being solved for and avoid over-fitting.  The procedure will increase the grouping for tilt angles, but the change in leave-out error is minor.

Tomogram Positioning

This sample can be positioned by viewing X/Z slices in the binned-down tomogram.

• Set Positioning tomogram thickness to 800.
• Press Create Whole Tomogram to build a whole, binned-down tomogram.
• Press Create Boundary Model to open the tomogram
• Open an XYZ window with Image-Xyz.  Set Sum in the toolbar to 20.
• Draw 3 pairs of lines: in the middle of the tomogram (Y ~380) and at Y values near 115 and 555.
• Save the model with the S key.
• Press Compute Z Shift & Pitch Angles.  The Added angle offset should be small; if it was more than a ~2 degrees, you would want to go back to the Fiducial Model Generation page and retrack with the new total angle..
• Press Create Final Alignment to get a new alignment that incorporates the new total angle offset and the Z shift.

Final Aligned Stack Creation:

• Press Create Full Aligned Stack.

Tomogram generation:

• Make sure only one of the two choices Use the GPU and Parallel processing is selected (unless you have multiple GPUs), and press Generate Tomogram.
• Press View Tomogram in 3dmod when it is done.

Post-processing:

• Turn on the rubber band in the Zap window and select a subarea area if desired.
• Scroll to the first slice that you want to keep and press Lo, then find the last slice you want to keep and press Hi.  You should be able to trim off most of the crevasses on the bottom.
• In Etomo, press Get XYZ Volume Range from 3dmod to import these limits.
• In the Zap window, select a narrower range of slices with Lo and Hi that will be used to determine scaling.
• In Etomo, press Get XYZ Sub-Area from 3dmod in the Scaling section to import these limits.
• Press Trim Volume.  When done, press 3dmod Trimmed Volume to see the result.

Clean Up:

Archive the original stack and clean up as usual.