EMAN Release Notes (1.4)
Welcome to EMAN 1.4. I'd strongly encourage you to take a look at the release
notes for this version before proceeding. Remember the full online help for
EMAN is always available by pressing the "HELP" button in eman.
EMAN 1.4 contains a significant number of new refinement options, several
of which may have a substantial impact on reconstruction resolution. This
does not invalidate previous results, it simply means you may be able to
take a data set that previously stopped at 11 A resolution, and achieve 9
or 10 A resolution with careful application of the new options. We have tried
to update the tutorial documentation to take the new options into account
to some extent, but it is likely to require some trial and error to achieve
optimal results for any specific project.
In addition to the new options, the full lecture notes and most of the
videotaped lectures from the EMAN workshop held in Dec 2002 are now available
online at:
http://ncmi.bcm.tmc.edu/ncmi/events/workshops/workshops_19
EMAN is now entering a transition period. While this release is quite similar
to the previous release, we are now in the midst of a major EMAN library
overhaul. The new EMAN library will be much more modular, with the goal of
making it more accessible to programmer/users out there. The next version
will have much improved Python scripting, and the library will be written
such that it should be extremely easy for end users familiar with C++ to
add new filters, alignment routines, comparison metrics, etc. This new version
will be dubbed EMAN 2.0 and should be released towards the end of this year.
Our hope is that this new EMAN release will embody more of the true 'open
source' nature of EMAN, and will allow others to develop new algorithms and
techniques within a robust (and free) platform. We will, of course, attempt
to maintain as much backward compatibility with EMAN 1.4 as possible.
Here's a list of the major changes in this version:
- 3dit/3dit2 - A number of recent problems have been associated with
these options, and we have found that they are actually not necessary in
many cases. If you experience strange problems with your 3D models containing
horizontal bands, or being completely black, try not using these options.
We are working to resolve this issue in a more satisfactory way. If they
are necessary for your project (try running clean3d manually on a model and
see if there is a substantial impact), look for an upcoming nightly release
version to address this problem.
- Substantial improvements were made to EMAN's parallelism. When
used on small particles on a large number of processors, the previous versions
of EMAN would run at less than 100% capacity on each processor. The new
version should run at 100% on fairly large numbers of CPUs.
- Python support - EMAN now officially supports Python scripting.
Some very powerful operations can be performed with some very simple (and
easy) scripts. Just to give you the flavor, here's a quick example which
will compare class-averages to projections:
python
from EMAN import *
img=readImages("classes.1.hed",-1,-1)
for i in range(0,len(img),2):
print i," vs. ",i+1," -> ",img[i].lcmp(img[i+1],1,0)[0]
New options in the refine command:
- setsf=<res> - Will apply the user defined 1D structure factor
to the final reconstructed model after each iteration. This represents the
ultimate in CTF amplitude correction. While the 'ctfcw=' option does a
very good job of CTF correction, it does make certain assumptions which
aren't valid in all cases. If you have a reliable structure factor (from
x-ray solution scattering, or generated from a set of micrographs), this
option is highly recommended.
- refine - How confusing, an option with the same name as the
command. Normal alignments in EMAN classification are +- 1 pixel. This option
will refine all 2D alignments to +-.05-.1 pixels. This can signficantly
improve classification accuracy for certain structures, and hence improve
resolution. In some cases it will have little or no effect. There is a significant
speed penalty associated with this option.
- phasecls - This option will use signal to noise ratio weighted
phase residual as a classification criteria (instead of the default optimized
real space variance). Over the last year or so, people working on cylindrical
structures (like GroEL), have noticed that 'side views' of this particle
seem to frequently get classified as being tilted 4 or 5 degrees from the
side view. While apparently this didn't effect the models significantly at
the obtained resolution, it is quite irritating. This problem turns out to
be due to resolution mismatch between the 3D model and the individual particles.
Using phase residual solves this problem, although it's unclear if there
is any resolution improvement. This option has a slight speed penalty
- fscls - An improvement, albeit an experimental one, over phasecls.
phasecls basically ignores Fourier amplitude when making image comparisons.
fscls will use a SNR weighted Fourier shell correlation as a similarity
criteria. Preliminary tests have shown that this produces slightly better
results than phasecls, but it should still be used with caution.
- slow - This option tries to do an even better job of 2D particle
alignment. Used with the 'refine' option above, this option uses fsc as
a criteria during alignment. This is quite slow, but theoretically even more
accurate.
- axialfilt=<val> - In theory, direct Fourier reconstruction
algorithms, such as the one used in EMAN are not susceptible to problems
with an excess of images in a particular set of orientations. However, there
are certain circumstances under which this is not the case, and a certain
amount of 'axial blurring' can occur. This largely occurs with cylindrical
particles where virtually all of the data is in side views. Axialfilt can
perform some empirical corrections for this effect.
qindex - a program for very basic indexing of 2D crystal images
was added. This program may be enhanced in future, depending on community
response.
fitctf - improvements in fitting quality and accuracy. May still
have problems with some CCD data.
procpdb.py - a script for basic manipulation of PDB files, designed
largely for use with foldhunter
foldhunter - new option added to generate rotated/translated PDB
files corresponding to the best fits
multirefine - now permits a different symmetry to be specified
for each model. This is still an experimental option.
building EMAN from source - We previously used GNU configure to easily
configure EMAN on any platform. Due to a variety of problems with this system,
we are now switching to 'CMake'. This is the cross-platform configuration
tool written as part of the ITK project. If you wish to compile EMAN from
source, please download and install the cmake version provided on the EMAN
required libraries download page. The INSTALL text file will describe how
to use it. GNU configure is still present, but is deprecated.
As always, comments, suggestions and questions are always welcome. Just
email sludtke@bcm.tmc.edu.
Good luck, and enjoy EMAN !
----------------------------------------------------------------------------
Steven Ludtke, PhD | Baylor College of Medicine
sludtke@bcm.tmc.edu | Co-Director
stevel@alumni.caltech.edu | National Center For Macromolecular Imaging
V: (713)798-6989 | Dept of Biochemistry and Mol. Biol.
instant messenger: sludtke42 | * Those who do ARE *
http://ncmi.bcm.tmc.edu/~stevel | The converse also holds