$0

EMAN Reconstruction - Phase 2


Building a preliminary model. This step varies considerably depending on the symmetry of your model. $csym

1. Cn symmetry - use startcsym

$csym.1

If you know your model has a Cn symmetry (n>2), or a Cn pseudosymmetry, it is generally very easy to generate an adequate preliminary model in EMAN. The command startcsym performs this task automatically from start.hed/img. This program will automatically search for particles which are top views (along the Cn axis) and side views of the macromolecule. It then generates a class average for both of these views, then builds a rough 3d model from the 2 class averages. This model will be VERY rough and noisy. This isn't generally a problem. In most cases this model will be sufficient for the refinement loop to converge. To use the program, type:

startcsym start.hed #-ptcl-to-use sym=cn

Replace n with the symmetry of your particle. '#-ptcl-to-use' is the number of particles used to generate the top and side views. This number should generally be between 20 and 100, and should not be more than 5-10% of your total number of particles. A larger number will give less noisy class-averages, but they will be blurred out by averaging a wider range of different orientations. A smaller number will give a sharper but noisier result. If you have a LOT of particles (>5000), you might want to use only the first few thousand for this step to save time, eg:

proc2d start.hed tmp.hed last=3999
startcsym tmp.hed 60 sym=c4
rm tmp.hed tmp.img

After running this command, several new files will exist:
cls000.hed - the individual particles used to generate the top view
cls001.hed - the individual particles used to generate the side view
classes.hed - contains 2 class averages, the top view and the side view
sym.hed - contains 3 images, the symmetrized top view, the side view, and the side view aligned to the symmetric axis
threed.0a.mrc - The preliminary 3d model.

You should take a look at classes.hed and sym.hed (use the eman file/history browser). Make sure that the symmetrized and unsymmetrized top views look somewhat similar. Make sure the side view isn't too ridiculous. If these are ok, chances are, the 3d model is good enough for the next step. Don't bother looking at the 3D model in projection. It will look terrible from most angles. An isosurface rendering will give you a better idea.

There is a possibility that the orientation of the side view may be determined incorrectly (the 3rd image in sym.img). Usually, if startcsym is wrong, it will be off by exactly 90 degrees. If this occurs, run:

startcsym start.hed #-ptcl-to-use sym=cn nosym fixrot=90

This will fix the 3D model withour rerunning the symmetry search. Double-check that it worked properly if you do this.

If the top view doesn't seem to have the correct symmetry, or the side view seems obviously wrong, then this technique may not work in your case. This problem can occur for one of several reasons. First, your particle may have a preferred orientation, and there may simply not be any top views present in the data set. Second, the image contrast might simply be too low for an accurate symmetry search (in this case, you might consider trying a higher dose, or negative stain for an initial model). Finally, the particles may not be well-centered in the box. The symmetry search can tolerate some degree of poor centering, but too much can make it fail.

If this method fails, you can try specifying 'unk' for the symmetry and following these instructions, but you might be stuck. EMAN has some routines for generating a rough tomographic reconstruction from a tilt series, but it is still experimental and undocumented (you can email me if you'd like to try it).

If this method worked properly, you might also want to consider low-pass filtering the result before using it for refinement. Typically, a low-pass filter to 20 or 30A will help make the model more suitable for refinement. You can do this with proc3d (just remember that the filter is specified in Fourier pixels, NOT Angstroms. $dsym

1. Dn symmetry - use startcsym

$dsym.1 If you know your model has a Dn symmetry (n>2), or a Dn pseudosymmetry, a preliminary model can be generated using the same method as Cn symmetry. The command startcsym performs this task automatically from start.hed/img. This program will automatically search for particles which are top views (along the Cn axis) and side views of the particle. It then generates a class average for both of these views, then builds a rough 2d model from the 2 class averages. This model will be VERY rough and noisy. This isn't generally a problem. In most cases this model will be sufficient for the refinement loop to converge. To use the program, type:
startcsym start.hed #-ptcl-to-use sym=cn

Replace n with the symmetry of your particle. Note that you specify C symmetry for this program, even if your particle has D symmetry. '#-ptcl-to-use' is the number of particles used to generate the top and side views. This number should generally be between 20 and 100, and should not be more than 5-10% of your total number of particles. A larger number will give less noisy class-averages, but they will be blurred out by averaging a wider range of different orientations. A smaller number will give a sharper but noisier result. If you have a LOT of particles (>5000), you might want to use only the first few thousand for this step to save time, eg:

proc2d start.hed tmp.hed last=3999
startcsym tmp.hed 60 sym=c4
rm tmp.hed tmp.img

After running this command, several new files will exist:
cls000.hed - the individual particles used to generate the top view
cls001.hed - the individual particles used to generate the side view
classes.hed - contains 2 class averages, the top view and the side view
sym.hed - contains 3 images, the symmetrized top view, the side view, and the side view aligned to the symmetric axis
threed.0a.mrc - The preliminary 3d model.

You might want to take a look at classes.hed and sym.hed. Make sure that the symmetrized and unsymmetrized top views look somewhat similar. Make sure the side view isn't too ridiculous. If these are ok, chances are, the 3d model is ok to try refining. Don't bother looking at the 3D model in projection. It will look terrible from most angles. An isosurface rendering will give you a better idea.

There is a possibility that the orientation of the side view may be determined incorrectly (the 3rd image in sym.img). Usually, if startcsym is wrong, it will be off by exactly 90 degrees. If this occurs, run:

startcsym start.hed #-ptcl-to-use sym=cn nosym fixrot=90

This will fix the 3D model withour rerunning the symmetry search. Double-check that it worked properly if you do this.

If the top view doesn't seem to have the correct symmetry, or the side view seems obviously wrong, then this technique may not work in your case. This problem can occur for one of several reasons. First, your particle may have a preferred orientation, and there may simply not be any top views present in the data set. Second, the image contrast might simply be too low for an accurate symmetry search (in this case, you might consider trying a higher dose, or negative stain for an initial model). Finally, the particles may not be well-centered in the box. The symmetry search can tolerate some degree of poor centering, but too much can make it fail.

If this method fails, you can try specifying 'unk' for the symmetry and following these instructions, but you might be stuck. EMAN has some routines for generating a rough tomographic reconstruction from a tilt series, but it is still experimental and undocumented (you can email me if you'd like to try it).

If this method worked properly, you might also want to consider low-pass filtering the result before using it for refinement. Typically, a low-pass filter to 20 or 30A will help make the model more suitable for refinement. You can do this with proc3d (just remember that the filter is specified in Fourier pixels, NOT Angstroms. $icos

1. Icosahedral symmetry - use starticos

$icos.1

Please note that, while EMAN does support icosahedral symmetry, the current version is better suited for lower symmetries. For particles with such high symmetries, other techniques may work better than the one used by EMAN. We have successfully used EMAN to refine several icoshedral particles, so it's useable, just remember that it's one of the more poorly supported features in the current version. We are actively working on optimizing these routines for future releases.

There are many ways to generate a preliminary model from icosahedral data. EMAN uses a particularly simple and fast routine that will work well in many cases. The program starticos searches the data set (start.hed) for particles with the best 5, 3 and 2-fold symmetries. It will locate a number of particles for each symmetry, then align/average them to make a class average for each symmetric axis. These 3 views are then used to generate a 3d model. This routine will generally work well as long as the distribution of orientations contains particles near all 3 symmetries. If one or more is under-represented in the data, the result may not be satisfactory. Use it like this:

starticos start.hed #-ptcl-to-use [imask=radius]

The #-ptcl-to-use should generally be between 10 and 100, and less than 5-10% of the total number of particles. imask is optional. If your particle has a lot of non-icosahedral mass in the center (like DNA/RNA), this option allows you to exclude the center of each particle from the symmetry search. This may produce better results in some cases. Specify the exclusion radius in pixels if desired.

After running this command, several new files will exist:
cls000.hed - the individual particles used to generate the 5f view
cls001.hed - the individual particles used to generate the 3f view
cls002.hed - the individual particles used to generate the 2f view
classes.hed - contains 3 class averages, the 5f, 3f and 2f views
sym.hed - Contains the symmterized class averages
threed.0a.mrc - The preliminary 3d model.

You might want to take a look at classes.hed and sym.hed. Make sure that the symmetrized and unsymmetrized views look somewhat similar. If these are ok, chances are, the 3d model is ok to try a refinement on. $asym

1. No symmetry (or C2 symmtery) - use startnrclasses and startAny

$asym.1

First, use startnrclasses. This program will group your boxed out particles into self similar classes (reference free classification), then align and average the particles in each class. This will give you a set of low-noise images representative of the various projections present in your data set. If you didn't do it earlier, process your particles with cenalignint before running:

startnrclasses start.hed #-Ptcl-per-class [mask=radius]

Generally the number of particles in each class should be at least 20 or so. Usually 40 or 50 classes will be enough to analyze, so the total number of particles divided by 50 is usually pretty good. The optional mask allows you to exclude information outside a given radius (in pixels). This number should be slightly larger than the longest axis of your particle. If you don't provide a mask, the box radius will be used.

Next, you should manually examine the class averages in classes.nr.hed. Select some class averages that look good, and put them in a file called good.hed. The precise number of classes to select depends on the particle. Generally at least 7 or 8 should be selected. You can take as many more as you like, but try to avoid particularly noisy ones, or duplicates of the same view. Once you have good.hed prepared, run:

startAny good.hed [sym=symetry]

If your particle has some symmetry, but was not amenable to the other methods, then specify it above. For asymmetric particles, don't specify sym=. $unk

1. Unknown symmetry - use startnrclasses and startAny

$unk.1

If you don't know the symmetry of your molecule, it's best to try to make a preliminary guess, then relax the symmetry later if it proves to be incorrect. startnrclasses is a useful tool for making this determination. This program will group your boxed out particles into self similar classes (reference free classification), then align and average the particles in each class. This will give you a set of low-noise images representative of the various projections present in your data set. This, often coupled with some biochemical knowldege, should provide a good idea of what the particle's symmetry is. First, run:

startnrclasses start.hed #-Ptcl-per-class [mask=radius]

Generally the number of particles in each class should be at least 20 or so. Usually 40 or 50 classes will be enough to analyze, so the total number of particles divided by 50 is usually pretty good. The optional mask allows you to exclude information outside a given radius (in pixels). This number should be slightly larger than the longest axis of your particle. If you don't provide a mask, the box radius will be used. You should run cenalignint on your particles before using startnrclasses if you haven't already.

Once startnrclasses has been finished (this may take some time), you will have a set of cls files which contain the particles split into the individual classes. You can ignore these for the most part. A file called classes.hed will also be produced. This contains the class averages you should examine with eman.

Examine these images. Many of them will be awful, and many will be similar to each other (in most cases). What you need to do now is try to locate a few of these class averages which look pretty good, and represent as wide a range of orientations as possible. That is, don't pick 2 that look almost the same, even if the noise level is low in both. You can generate a new file with only the 'good' images in them as follows:

Simply close the 'Big View' when you're done with it. You might want to examine good.hed in the browser, just to make sure it contains all the images you wanted in the correct order.

The next step is to take these images and try to build a 3D model from them. This is accomplished with the 'startAny' command, as follows:

startAny good.hed [sym=c<n>] [proc=<nproc>]

Specify symmetry if it is known, but only a single c symmtery is allowed in this program. This will run for some time. It uses cross-common lines in Fourier space to try to determine the relative orientations of the particles, then eventually builds a 3d model. As with the other startup programs, the output will be written to threed.0a.mrc. When it's done, you might want to have a look at this model in projection or as an isosurface, to see if it came out well. To look at it in projection, either select it in the eman file browser, and rotate it with the right mouse button, or use v4 threed.0a.mrc. For isosurfaces, use your favorite MRC capable rendering program. Vis5d is included with the EMAN package. It's an excellent (and free) package originally designed for weather visualization. To use it, type:

mrc2v5dt threed.0a.mrc x.v5d
vis5d -box 1 1 1 -path / x.v5d

Read the vis5d reference manual referenced above for more info. $2

2. Go on to step 3

You should now have start.hed/img and threed.0a.mrc. The hardest part is over. This is what you need to start a refinement. Go on to step 3.