Miscellaneous procedure files



Projection onto convex sets

Projection onto convex sets (POCS) uses a priori information to adjust a reconstruction. Specificially in this case, the SPIDER operation 'CV REPL' uses information about the missing cone (or wedge), coupled with knowledge about the approximate envelope function. One can thus rule out structures that extend beyond the mask, and you can place some constraints on the data in the missing cone.

Use of POCS has 2 required steps and 1 optional step:

  1. Generate a 3D binary mask to use an an envelope for your particle

    2. The procedure file mask3d.rct will threshold Volumes/vcla*** and make a binary mask.
    Outputs:

    ¤ vcla***-mask: Final, binary, 3D mask.
    ¤ vcla***-masked: Input volume upon multiplication by the mask.
    Check to make sure that nothing of interest is cut off by the mask.
    ¤ tmpmask3_mask_filt: Intermediate mask before final threshold (x13) is applied. See below.

    Parameters:
    ¤ filter radius, x11: The goal is the make a generous, smooth mask that doesn't cutoff any part of the structure that might possibly be real.
    The filter radius (used twice) first is applied to your reconstruction and then to the intermediate binary mask to smooth it further.
    Units are in Angstroms.
    ¤ input volume threshold, x12: You want a cutoff that your structure will exceed but that the background won't.
    The volume average is a decent starting point, but there will be some trial and error involved.
    Check the statistics of your input volume, either by running this procedure file, or with the 'FS' operation.
    ¤ intermediate mask threshold, x13: The first binary mask will be jagged, so it is filtered again and thresholded.
    To minimize the trial and error, the filtered intermediate mask, tmpmask3_mask_filt is saved,
    so that one can find an appropriate threshold without trial and error in Chimera.

  2. [Optional] Choose a projection where the missing-cone artifact looks strong

    3. The procedure file refproj.rct will generate set set of projections from Volumes/vcla{***[single-class]}. Set the parameter [single-class] to 0 to project all class volumes. Outputs:

    ¤ prj_***: Stack file of re-projections.
    ¤ stkpws_***: Stack of power spectra of re-projections.
    ¤ refangles: Euler angles for each projection.

    Look at the montage of the projections, and pick a projection where the missing-cone artifacts look particularly strong. The output stkpws_*** shows power spectra of the re-projections, and for high values of the Euler angle theta, the missing cone will be obvious. Then, note from the doc file refangles the 3rd Euler angle, phi, for that projection. For example, I chose projection #77, which corresponds to phi = 65.4 degrees. In the next step, we will monitor this projection over each iteration of POCS.

  3. Run projection onto convex sets

    4. The procedure file pocs.rct runs a number of iterations (specified by parameter [num-iter]) of projection onto convex sets. The input volume Volumes/vcla*** will be constrained by the envelope specified by Volumes/vcla***-mask. Outputs:

    ¤ Pocs/yprj_***_###: Projections for each iteration of POCS, in the direction ([yprj-phi],90,0). More below.
    ¤ Pocs/volpocs***: Reconstruction after POCS.
    ¤ Pocs/volpocs***-filt: Low-pass filtered reconstruction.

    Parameters:
    ¤ [num-iter]: Number of iterations of POCS.
    In my test case, the projection didn't change much beyond 50 iterations, but I requested 99.
    ¤ [tilt-angle]: Tilt angle during the data collection.
    In my test, I erred on the side of high tilt, wanting to minimize the region where POCS adjusts the missing data.
    My actual range was 42.6 to 50.1 degrees, and I used 50 in the procedure file.
    ¤ [yprj-phi]: Euler angle phi for projection recorded at each iteration of POCS.
    In the previous (optional) step refproj.rct,
    you could choose a view where the missing-cone artifacts look particular strong.
    In my example, projection #77 was such a view, which corresponds to an angle of 65.4 degrees for Euler angle phi.
    A projection in that case would be taken at Euler angles (65.4,90,0) for each iteration.

    Other procedure files

    Source: misc.htm     Page updated: 2015/08/13     Tapu Shaikh