EMAN 1.6 (with Windows support)

[Steve Ludtke]

Hi. We're happy to formally announce the availability of EMAN 1.6. While
this has been available for several weeks, the release has now achieved a
reasonable level of stability. An important note to many of you is the
availability of the first (experimental) Windows XP binary release. In
theory this release includes full GUI, and even python scripting support.
In practice, the GUI tools work, but nobody has yet tried any sort of
reconstruction under Windows. We will provide only limited support for
the Windows binary. Do NOT expect to use this for parallel refinements.

Major changes in this version include:
- Inclusion of (experimental) EMANimator and AIRS modules for Chimera in
  the binary releases
- dfilt option in refine works now. Note that this option works best,
  only in conjuction with 'usefilt', and may be tricky to optimize.
  'phasecls' or 'fscls' may be better for routine use. This is the major
   cause for the 1.6 release. The (fixed) dfilt option was used for our
   recent 6 A reconstruction of GroEL, and we have been touting it as the
   best current approach for doing EMAN reconstructions (which it is).
   Unfortunately, due to an error on my part, a critical
   change never got updated in the EMAN source repository. So,
   essentially, this option worked very well for me, but not for anyone
   else. Anyone who was experimenting with the dfilt option
   should give it another try. On the bright side, the bug wasn't causing
   any harm, specifying dfilt in the buggy version was almost equivalent
   to not specifying it at all. I apologize for this little snafu.
- Better support for icosahedral reconstructions and for
  reconstructions of large models (contact wjiang at bcm.tmc.edu for more
  info). EMAN has now sucessfully produced a subnanometer resolution
  icosahedral virus structure.
- Some programs didn't like EMAN's MRC files. They should now be properly
  CCP4 compliant.
- EM file format support was minimal in last version, now properly handles
  stack files and byte-ordering issues
- New program 'qplot'. Utility for plotting 2 column text files.
- EMAN can now compile and run on Macs (again). OS 10.3 only, with
  appropriate fink packages installed. This is still not officially
  supported.
- The >2G file support was apparently broken in v1.5. This has
  been fixed.
- Improved 64-bit support for use on Athlon-64's and Opterons. This
  is currently the fastest platform available for EMAN refinements
  (and the most cost-effective by a fair margin)
- Support for KDE 3.2/Mandrake 10.0
- Fixed funny narrow display in some cases in asymmetric triangle viewer
- fitctf improved to work on a wider range of images, including particles
  from CCD frames
- usefilt support in multirefine


I've been hearing an increasing number of statements from people about
EMAN being slow. I'd like to clarify a bit. When used with the correct
options, EMAN can produce low resolution structures very quickly, even
compared to other available packages. However, when all of the new
features designed to improve the veracity and resolution of the
reconstruction are enabled, it does, indeed, take a lot of CPU time to
run. These changes were necessary to go from 11 A to 6 A resolution on
GroEL. There is still room for optimization and improvement, but at the
moment, we are focused more on achieving the best possible structure from
the available data and less on doing it quickly. When used properly, EMAN
can virtually eliminate problems with initial model bias, which often
improperly exaggerate resolution measures, while producing poor quality
models.

One note of caution is called for however. I've noticed that a
number of users have recently started using some of the more 'aggresive'
refinement options in particularly risky combinations. EMAN has a number
of features designed specifically to avoid problems with initial model
bias and algorithmic bias. However, it is possible to turn off or
substantially degrade these features through use of specific refinement
options. While this can be a good thing, and lead to higher
resolution refinements, a great deal of care must be taken to insure that
observed features are really represented by the data. I strongly suspect
that a number of the apparent 'conflicts' that have been appearing in
recent publications (with various software pacakages), are
due to problems with model/algorithm bias. Among the important tests to
run on novel structures is to rerefine your model from a starting model
with an intentionally added  artifact to insure that it is eliminated
through refinement, or even to refine from completely different starting
models (for example, a featureless ellipsoid). The more 'aggressive'
options of the refine command include: 'classiter=0', 'setsf=',
'amask=(with very tight masks)'.


----------------------------------------------------------------------------
Steven Ludtke, PhD              |        Baylor College of Medicine
sludtke at bcm.tmc.edu             |       Asst. Professor & Co-Director
stevel at alumni.caltech.edu       | National Center For Macromolecular Imaging
V: (713)798-9020                |    Dept of Biochemistry and Mol. Biol.
F: (713)798-1625                |
instant messenger: sludtke42    |             Those who Do, Are
http://ncmi.bcm.tmc.edu/~stevel |         The converse also applies