A graphical program for locating particles within micrographs.

Usage boxer [<filename>] [area=<num>,<denom>,<box size>]


Boxer is a graphical program for selecting particles from micrographs (or CCD frames). Particle selection can be performed manually or semi-automatically. Run the program with the name of the image containing particles to be selected.

Machines with limited memory

Since these images are often very large, possibly larger than your machine's memory, the 'area=' option is provided to allow only a portion of the image to be examined at once. Currently this option only works with images in the MRC format (Without this option, boxer can read images in any of the supported file formats). The 'area=' option has 3 parameters. The second parameter is the number of regions to split the micrograph into. This number must be an integer squared (ie - 4,9,16,...). The third number is the particle box size. When individual sections of the micrograph are read, enough overlap between sections is allowed to insure that particles near the edge of the secion can still be selected. Finally, the first number specifies which area should be read. This value is ranges from 0 to n-1. That is, if the second number is 4, the first number is 0,1,2 or 3. You can run 'iminfo <image file>' to determine how much memory the entire micrograph will take. Generally you want to make sure you have at least 50% more memory than this. If you have less, you should use the 'area=' option or find a machine with more memory.

Using boxer

The control panel for boxer is shown below.

A. Measure Mode - The mouse can be used to measure distances in the micrograph B. Select Mode - The mouse can be used to select particles and/or drag existing boxes C. Delete Mode - The mouse can be used to delete particles in either view D. Draw Mode - The mouse can be used to blank out areas of the micrograph containing contamination to prevent autoboxing E. Box Size - The box size in pixels for Select and Draw mode F. A/Pix - Angstroms/pixel used in Measure Mode G. Scale - The scale used in the image display H. Panner - This allows you to pan around the micrograph I. Helix Mode - Special helix boxing mode allows filamentous particles to be extracted as a set of particles J. Helix Submode - Allows for some subtle modifications to the helix boxing mode K. Overlap - Specifies the amount of overlap between boxes in helix mode

There are 3 windows in boxer. The control panel, the micrograph view, and the particle view. The control panel is described above. The micrograph view will display some fraction of the overall micrograph. The 'scale' in the control panel determines the scaling factor for this window. To pan around the micrograph, either use the Panner (H) in the control panel, or drag directly in the micrograph view with the middle mouse button. The particle view contains the individual boxed out particles. In select mode, you can click on a particle in the particle view and that particle will be centered in the micrograph view. In delete mode, particles can be removed from either the micrograph or the particle view.

To adjust the brightness and contrast of either the image view or the particle view, click the center mouse button anywhere in the image. This will cause an 'image inspector' window to appear. This contains a variety of adjustments, and even allows you to save snapshots of the window to use in web pages or reports.

Semi-automatic Boxing

Particles can be semi-automatically selected. This is done with a simple correlation based method. There are four versions of this feature. Two are currently functional, one for use on new particles where no 3D model yet exists, and another for cases where a low-resolution 3D model is available. This feature is relatively easy to use, and generally works reasonably well, especially when a 3D reference is available:

No 3D model available:

  1. If there are large areas of contamination, or the edge of the holey film, etc., consider using the 'Draw' mode to paint out these areas. Otherwise you may have to manually delete any boxes incorrectly assigned to these areas.
  2. Manually select a few representative particles in 'select' mode. Try to select a diverse set of particle orientations. In most cases 2 or 3 particles is sufficient. The more particles you select here, the slower the autoboxing will be.
  3. Select 'Autobox' from the 'Boxes' menu. In a few moments, you should notice the list of boxed particles will change, and another window will appear with 4 sliders. As you adjust the sliders, a 1024x1024 area surrounding the first particle you selected manually will be continually reboxed. These sliders represent several thresholds which are used to decide what is and isn't a particle. Since boxing the entire micrograph takes some time, these sliders are adjusted with only a 1k x 1k area being interactively reboxed. This allows accurate threshold setting for optimal autoboxing.
  4. First, adjust the top slider until everything that could possibly be a particle is selected. Then adjust the next three sliders to eliminate anything that isn't a particle (you might not manage to get rid of all of the non-particles). Generally, it's best to slide the last 3 sliders as far to the right as possible without eliminating any real particles. When you've got the sliders adjusted to do a pretty good job in selecting the particles in this region of the micrograph, press 'OK'. The remainder of the micrograph will then be automatically boxed.
  5. You may wish to manually prune the particles in delete mode when the autoboxing is complete. IF you did not use 'Draw' mode, this will be especially necessary.
  6. If you find that the autoboxing worked very poorly, you might want to try again with different threshold settings. To do this, select 'Clear boxes' from the 'Boxes' menu, then start again on step 2.

When you've got all of the particles in the micrograph selected, you can save the selected particles using the 'Save Boxed Particles' command on the Boxes menu. You may also wish to save a box database file (containing a list of box positions in the micrograph). If you ever wish to go back to the original micrograph and modify the box positions, you'll need this file. This is saved using the 'Save Boxes' option on the Boxes menu.

Preliminary 3D model available:

The basis for this technique is preparing a set of reference particles from the available 3D model. Unlike the above method, no rotational averaging is performed, making the particle matching much more accurate. However, this means each reference must be prepared over a range of angles of in-plane rotation. Since this would produce a very large number of references which would make automatic boxing take prohibitively long, a novel procedure for determining an optimal smaller set of references is used. Once the reference set has been prepared, they may be reused. A script, new in version 1.2, is provided to make preparing the references easier. The script performs the following tasks. Note: you do NOT need to perform these tasks manually :

  1. project3d <preliminary model> sym= prop=30 phitoo

    • Specify the symmetry if the model has any. You may wish to adjust the angle (prop=) depending on the symmetry. Keep in mind that this angle is used for in-plane rotations as well, so a small number can produce a lot of projections.
  2. classesbymra proj.hed proj.hed matrix split norot
    • This process compares every projection with every other projection and stores the results in matrix.dat.
  3. mx2img matrix.dat matrix.mrc
  4. matrixembed matrix.mrc self.loc euler=proj.hed bestref=50
    • This determines the best set of reference images and puts them in best.hed. 'bestref=' specifies how many references to prepare, however, the images in best.hed is sorted. So, the best n references will always be the first n images in best.hed.
  5. Decide how many references you want to use. More references will mean slower, but more accurate autoboxing. To do this, use eman to examine best.hed and decide at what point this file can be truncated. Assuming the projections have the opposite contrast of the micrographs, the particles must be inverted as well. If, for example you decide to use 15 references, do:
    • proc2d best.hed boxref.hed last=14 invert

To do all of this in one step, simply run: <threed file> [sym=<sym>] [nref=<n to gen>] [invert] [ang=<dalt>] [phi=<dphi>]

Generally you don't want to specify ang= or phi=, the automatic values will suffice. EMAN uses higher densities as positive values by default. Generally electron cryomicrographs will have smaller values for higher densities. If this is the case for you, then you will want to specify the invert option, so the reference image contrast matches your micrograph images. nref specifies the number of reference images to generate. For an asymmetric particle, you may want as many as 40-50. With some symmetry, you may be able to drop this. Generally you'll pick a larger value than you expect to use, then examine the results manually with eman. What you're looking for is a set of projections that represents all possible unique views of your particle at some unspecified level of detail. Generally more references will produce more accurate boxing, but will (obviously) take a proportionally longer time. Note that when selecting a subset of particles to use, you may truncate the list of references, but you should NOT randomly select particles from the list. The particles are sorted in order of their mathematical dissimilarity.

Now that you've got a set of reference images in boxref.hed/img, you're ready to autobox some micrographs:

  1. If there are large areas of contamination, or the edge of the holey film, etc., consider using the 'Draw' mode to paint out these areas. Otherwise you may have to manually delete any boxes incorrectly assigned to these areas.
  2. Center an area of the micrograph where some particles are visible on the screen.
  3. Use the 'Boxes->Autobox from references' menu item, and select the boxref.hed file you created.

  4. A small area of the micrograph will be interactively autoboxed, and a panel with 4 sliders will appear. In this method, only the top 2 sliders are currently used. To get optimal thresholds, slide the second slider all the way to the left, then move the top slider to the left in small steps until everything you think is a particle is selected. Then, gradually move the second slider to the right until most of the improperly selected particles have been eliminated. If you cannot eliminate virtually all of the false positives with the second slider, you may have to use the first slider to get a better set. When the sliders are set, click OK.
  5. The entire micrograph will then be autoboxed. When the process is complete, you can (optionally) go through the particles manually and remove any remaining bad particles. When you're done, save the particles AND the box database.

Alternatively, there is an 'Boxes->Autobox from references (with SNR)' menu item. This uses slightly different criteria for particle selection: the first threshold uses a minimum estimated SNR to find all candidate particles, the second eliminates boxes with high residual noise energy, and the third eliminates boxes with high spectral variation in the residual noise. (The second and third have similar purposes, but one may be more effective in particular cases.) Here is one way of determining a good set of thresholds: (1) Set the SNR treshold to zero; (2) adjust the second and third sliders to eliminate most junk; and (3) fine tune the SNR treshold. Finally, it is strongly recommended that the user specify an accurate set of references when using this option (in particular, accounting for the CTF if possible).

Focal Pairs One common techniqe for locating particles in close to focus micrographs is to take a focal pair, and begin by locating the particles in the far from focus micrograph, where they are easier to find. Then the particles from the close to focus micrograph can be determined by mapping the boxes from one micrograph to the other. There are 2 ways to accomplish this mapping.

The first is somewhat manual, but works well in most cases. Locate the particles in the far from focus micrograph as described above. Then save the box positions to a file with the 'Save Box DB' menu item. Next, run boxer with the close to focus micrograph. Then read in the box database you just saved. Next you must manually adjust the positions of a few boxes in the micrograph. It's best to do this for at least one particle in each corner of the micrograph. Simply put the mouse in select mode and drag the boxes to their correct location. Note that it is VERY important to only do this for particles you can accurately adjust. Once you've moved a particle, it must be moved to the correct location. You cannot put it back where it started. If you make a mistake, clear the 'Clear Boxes' menu item and read the database in again. Once you've manually adjusted several box positions, select the 'Adjust Boxes' menu item. The other boxes will then be adjusted. If you did a good job manually moving the boxes, you should have a pretty good match now.

The second method for adjusting the box locations is fully automatic, but it requires that your computer have a lot of memory (enough for 2 micrographs at once), and it occasionally makes mistakes. For this method, box out the particles in the first micrograph as before. It's probably a good idea to save the particles and the box database for this micrograph before proceeding. Next, select the 'Focal Pair Autoalign' menu item. This will pop up a brief set of instructions. When you click OK, a file selector will appear. Select the file containing the close to focus micrograph. Boxer will then automatically align 1k x 1k areas of the first micrograph to the second micrograph. When it finishes, all of the box positions should be correct. You can now save the new box database and the boxed particles.

Iterative Centering

Sometimes, especially when using the first autoboxing method, the selected particles will not be well centered in their boxes. (In most cases, if the second autoboxing method is used, the centering will be quite good, and this procedure is not necessary). Previously, a program called cenalignint was often used to better center the particles when a problem did exist. The 'Iterative Centering' option in boxer mimics this procedure, but with several advantages. First, when centering already boxed particles, the edge of the particles would be filled in with zero when the particle was translated. This could lead to alignment problems later in processing. Second, this technique led to lots of stray files laying around. Third, there are no problems with particles being too close to the edge of the box and getting chopped in half.

Using this option in boxer will attempt to center the box positions as described in the cenalignint manual page. It does not work on all particles. For example, particles like GroEL, with very non-rotationally symmetric features in some views, work very poorly with this procedure. Using this technique on some particles will cause the boxes to become severely mispositioned. It is a good idea to save your box database before trying this option.

Normalize In the previous versions (before 1.2 (42)), boxer would automatically normalize the micrograph after reading it in. This no longer happens. A menu item is available if you wish to do this. Generally it's a good idea, unless you have a specific reason not to.

Stripe Filter Some scanners (most notably Zeiss, now Z/I Imaging) produce noticable vertical stripe patterns in scanned images. This is due to poor averaging when normalizing the CCD. This problem has been improved in recent versions of the scanner, I'm told, but still exists for many existing scanners. This filter will average the center 1/2 of the micrograph vertically and subtract the resulting values along vertical lines. This will remove the stripe artifacts in most cases. However, if there is strong contamination, or if you have very large high-contrast particles, this technique may actually produce new vertical stripes. Sometimes it's useful, sometimes it's not.

Median Filters

Another way to find particles in a close to focus micrograph with poor contrast is to filter the micrograph before locating the particles. Once the box positions are known, the data can then be boxed out of the unfiltered micrograph. The 'Process' menu allows you to apply several different median filters to the raw micrograph in memory. Once you've selected the particles in the close to focus image, save the box database, read the original image back in, and read the box database from the file.

Boxing filaments/helices

Selecting helices is fairly straightforward. When the mouse is in helix mode, boxer will work as in 'select' mode, except pairs of boxes define the endpoints of a filament. That is, click once on one end of the filament, then click a second time one the other end of the filament. Boxer will create all of the particle boxes along the filament length. You can adjust the position of either endpoint. You may create as many such helical segments as you like within a single micrograph. The 'olap' parameter allows you to select how much overlap there should be between boxes. By default, the boxes will be separated by 'boxsize'-'Olap' pixels measured along the length of the helix.

There are several submodes for helix mode, which allow other useful operations to be performed. In 'normal mode' the boxes are created along a straight-line path connecting the endpoints, and the 'particles' are simply boxed out normally. In 'Rotate Helix' mode, each individual box will be rotated so the helical segment will be aligned vertically in the box. This assumes, of course, that the filament is straight, and the endpoints are properly centered.

A further refinement is the 'Align Helix' mode, which will attempt to follow the path of the helix between the 2 endpoints. The helical segments are then rotated to be vertical based on a local tangent to the helical path. This option is still experimental (1.2 (42)), and may not work well in all cases.

The final mode is 'Unbend' mode. This mode will carefully trace the path of the filament between the endpoints. The boxed out 'particles' are no longer simply rotated, but are now unbent utilizing tangents and normals to the helical path at each pixel location. Of course this cannot compensate for true structural distortions present when a helix is bent, but it should be better than doing nothing at all. This option does not yet work properly in version 1.2 (42), but should work in subsequent releases.

Other Options

Most of the other options are experimental at this point. The 'check for drift' option is amusing, if not exceptionally useful. Once particles have been selected from the micrograph, this option will do a rough determination of the relative amount of drift present in various areas of the image. Boxer will calculate the 2D power spectrum of particles within a small area around each particle. The averaged power spectrum is then checked for asymmetry. The degree and direction of the asymmetry are used as an indication of the relative amount of drift. This will be displayed by drawing a line for each box. The length of the line indicates the relative amount of drift. Note that this value is relative. A micrograph with very little drift will still have long lines in certain areas.

Special Notes

Don't use too long path for the image files. If the whole path plus file name is over 1024 charaters, it will make this application crash.