AutoAlign Home

The script 'autoAlign' is a wrapper used to call 'CAN,' a program implementing a topology representing network ^[1] and multi-reference-alignment (M-R-A) using IMAGIC^[2]. Check the path to the IMAGIC executables to ensure that script will function properly (IMAGIC updates commonly change the inputs & order of inputs).

Prepare particles

Bandpass filter & normalize particles before running autoAlign:

    ** INCPREP (vs. 6-Aug-2010) (ACC.) welcomes you **
   
  Use MPI parallelisation [YES]                      : 
  Number of processors to be used [4]                : 
  Input file, image loc#s                            : all
  Output file, image loc#s                           : all_bpf
    
    The image will be band-pass filtered.
    Please specify:
  Low frequency cut off [0.05]                       : 
  Remaining low-freq. transmission [0.005]           : 
  High frequency cut off [0.8]                       : 0.7
    
   The image will be masked by a circle. Please specify
   the mask radius (pixels or fraction of inner radius)
   If you specify a drop-off it will be a soft mask.
  Mask radius, drop-off [0.8,0.0]                    : 0.7,0.2 
  Desired new sigma [10.0]                           : 
  Invert the image densities [NO]                    : 

Run autoAlign

Note that we have CAN_source to compile CAN on a single machine or computer cluster. The starting command is the same for each version:

autoAlign <# iter> <starting im> <orig im> <num particles> <starting class number> <ending class number> <filterstrength>

  <# iter> - number of iterations, typically 8 - 10.
  <starting_im> - starting particles
  <orig im> - original particles.  These are the same as starting particles unless you are inputting already rotated & shifted particles.
  <num particles> - used to set up neural network 
  <starting class number> - number of classums in first iteration procedure.  These correspond to 'nodes' in the neural network generated by CAN. Depending on sample, this number starts somewhere around 30 - 50 particles/class.
  <ending class number> -  From the starting class to the ending class number, autoAlign will incrementally decrease the number classums.  This last classum is usually around 200 particles/class.
  <filterstrength> - Spatial frequency applied to particles at each step of the iteration

Outputs:

References

1. ^ Ogura et al. 2003 Topology representing network enables highly accurate classification of protein images taken by cryo electron-microscope without masking. J. Struct. Biol. 140(3): 185-200 pmid=14572474
2. ^ IMAGIC: http://www.imagescience.de/