refine

• This is the main refinement loop in EMAN. It has a tremendous number of options.

Usage

• refine <total iter> mask=<rad> [proc=<maxproc>] [hard=<maxpr>] [simple] [ctfc=<res in A>] [ctfcw=<SF file>] [phaseopt] [setsf=<lowpass res>[,<highpass res>[,<SF file>]] [median] [nweight] [sym=<c2,c4,etc>] [ang=<dang>] [anglist=<fule>] [maxshift=<rad>] [pad=<size>] [classkeep=<sigma coef>] [classiter=<iter>] [refmaskali] [filt3d=<lp rad>] [sep=<n>] [xfiles=<a/pix>,<mass in kd>,<ali to>] [3dit=<n it>] [3dit2=<n it>] [speed=<1-5>] [sigfilt] [euler2=<oversmp>] [tree=<2,3>] [imask=<rad>] [amask=<r>,<thr>,<n iter>] [planarmask=<thickness>] [goodbad] [slow] [refine] [shrink=<n>] [projbatches=<n>] [ca3] [classfp=<msa#>] [perturb] [usefilt] [rfp] [precen] [phasecls] [fscls] [axialfilt=<res>] [msa=<proj>] [mra2] [3dr[=<range>]] [d3s=<sigmult>] [collapse=<ang>] [dfilt] [frm2d[=<rad>,<samp>]] [classalign2log] [continue]

Parameters

• <total iter>	Total number of iterations to be run in the current directory
  mask=<rad>	Applies an outside mask at several stages of refinement
  [proc=<maxproc>]	Number of proceesors to use for parallel processing
  [hard=<maxpr>]	Hard limit for <a href=make3d.html>make3d</a>
  [simple]	Use 'make3dsimple inorm'
  [ctfc=<res in A>]	Performs CTF correction filtered to the specified resolution (see <a href=classalignall.html>classalignall</a>)
  [ctfcw=<SF file>]	Performs CTF correction with Wiener filtration using a structure factor specified in a file
  [phaseopt]	Experimental modification to class-averaging
  [setsf=<lowpass res>[,<highpass res>[,<SF file>]]	Applies a final filter to set the radial structure factor to match the ctfcw structure factor multiplied by a gaussian filter. Filter is specified in A.
  [median]	Specify this when CTF correction is NOT being performed
  [nweight]	Used when start.hed contains averages, to properly weight during further averaging. This will not work in conjunction with median or ctfc/w.
  [sym=<c2,c4,etc>]	Imposes symmetry on the model, omit this option for no/unknown symmetry
  [ang=<dang>]	Angular step for projections (in degrees, see 'prop' in <a href=project3d.html>project3d</a>)
  [anglist=<fule>]	Use Euler angles included in this file for projection
  [maxshift=<rad>]	Maximum translation during image alignment
  [pad=<size>]	Pads the model during Fourier reconstruction (see <a href=make3d.html>make3d</a>)
  [classkeep=<sigma coef>]	Classkeep value for <a href=classalignall.html>classalignall</a>
  [classiter=<iter>]	Iterations for <a href=classalignall.html>classalignall</a>
  [refmaskali]	This passes the refmaskali option to classalignall if the amask option is being used, otherwise refmask is used when amask is specified.
  [filt3d=<lp rad>]	Apply a lowpass filter to the model after each iteration, this occurs before 3dit
  [sep=<n>]	This option is passed to <a href=classesbymra.html>classesbymra</a>
  [xfiles=<a/pix>,<mass in kd>,<ali to>]	Used to generate 3d models that are easy to evaluate
  [3dit=<n it>]	After constructing a model with Fourier reconstruction, this runs n iterations of real-space model 'cleansing' This option isdeprecated, suggest not using it
  [3dit2=<n it>]	This is real-space refinement with a special noise elimination feature enabled. This option is deprecated, suggest not using it
  [speed=<1-5>]	Speed parameter in clean3d (3dit,3dit2). Convergence rate, 1 dfl
  [sigfilt]	Applies a 1-sigma real-space filter to images at various points
  [euler2=<oversmp>]	Reassigns Euler angles to class averages with an optionally reduced angular increment. ie - 2 will use ang/2 (see <a href=classesrefs.html>classesrefs</a>
  [tree=<2,3>]	Faster classifications for large numbers of projections (see <a href=classesbymra.html>classesbymra</a>. Limited parallel support, see text.
  [imask=<rad>]	Applies an inside mask in several stages of refinement

  • [amask=<r>,<thr>,<n iter>]

    This applies an automask between steps in a reconstruction. see automask2 in <ahref=proc3d.html>proc3d</a>. REQUIRES xfiles to be used.

  [planarmask=<thickness>]	Masks the object along the z-axis with a total thickness specified in pixels.
  [goodbad]	Saves good and bad class averages from 3D reconstruction. Overwrites each new iteration.
  [slow]	This will use a very slow, exhaustive search, algorithm for 2D alignments. VERY SLOW
  [refine]	This will do subpixel alignment of the particle translations for classification and averaging. May have a significant impact at higher resolutions.
  [shrink=<n>]	Experimental, shrinks images at several points for faster runs
  [projbatches=<n>]	Experimental, split the projection into n batches to minimize total memory requirement
  [ca3]	Experimental. Uses classalign3 instead of classalign2
  [classfp=<msa#>]	Experimental, classification using rotational/translational invariants
  [perturb]	Experimental option to produce a more uniform distribution of points in 3D Fourier space.
  [usefilt]	Experimental
  [rfp]	Experimental, classification using rotational footprints
  [precen]	Indicates that raw particles have been accurately precentered (experimental)
  [phasecls]	Uses weighted mean phase error for classification (experimental)
  [fscls]	Uses FSC for classification (experimental)
  [axialfilt=<res>]	See proc3d, 0 provides maximal effect.
  [msa=<proj>]	Experimental
  [mra2]	Experimental
  [3dr[=<range>]]	Experimental
  [d3s=<sigmult>]	Experimental
  [collapse=<ang>]	Experimental
  [dfilt]	Experimental
  [frm2d[=<rad>,<samp>]]	This will classify using Fast Rotational Matching 2D. rad is the maximum radius of particles; samp is the number of angular sampling points at each radius circle
  [classalign2log]	Experimental. Calls classalign2 with classalign2log option to print out particle that were used in the class average.
  [continue]	If a job is killed in the middle of an iteration, this will allow it to pick up where it left off. Otherwise the last iteration will start from the beginning.

  Typical usage:
  • refine 5 ang=3 mask=28 pad=96 hard=25 classkeep=.3 classiter=8 proc=12 sym=d7 ctfc=12 sep=2 xfiles=4.2,1000,40

Description

Note: Despite the experimental tags, the 'phasecls' and 'fscls' almost always produce better results than the default linear variance classifier. Note that these options are mutually exclusive.

This is the main refinement command in EMAN. You should have your starting images in 'start.img' and your initial 3D model in threed.0a.mrc. The 'refine' command will then refine this model iteratively. Note that the 'total number of iterations' specifies the number of iterations the program will finish with. If 5 iterations have been run, then 'refine 8 ...' is run, the program will pick up after 5 iterations and run only 3 more cycles. If you want to start over again from the first iteration, you must remove all of the 'threed.?.mrc' files (except for 0a) and all of the 'classes.?.mrc' files.

Most of the options are simply passed through to other programs. Typical options for this program can be determined by running [/eman/eman1/progs/eman.html eman] and following the reconstruction procedure it provides.

The tree option may not work in certian parallel configurations, such as grid configurations where the filesystem isn't cross-mounted.

'frm2d' better used in combination with maxshift, which indicates the center of mass uncertainty.

EMAN Manual page, generated Mon Jan 8 17:35:59 2007