Incorrect pixel size in heterogenous refinement after unbinning

3D Refinement
1

I performed heterogenous refinement on particles in a 320 pixel box binned to 100 pixels (apix = 3.456 Å). I then reëxtracted selected particles binning to 160 pixels (new apix 2.16 Å), but a subsequent heterogenous refinement hit Nyquist at about 5.5 Å resolution
image

The particle metadata for that job shows:

      "passthrough": "particles",
      "num_items": 1271808,
      "summary": {
        "alignments_class_0/psize_A": 2.700000047683716,
        "alignments_class_1/psize_A": 2.700000047683716,
        "blob/shape": [160, 160],
        "blob/psize_A": 2.1600000858306885,
        "blob/sign": -1,
        "ctf/type": "spline",
        "ctf/accel_kv": 300,
        "ctf/cs_mm": 2.700000047683716,
        "ctf/amp_contrast": 0.10000000149011612,
        "ctf/phase_shift_rad": 0,
        "ctf/scale": 1,
        "ctf/shift_A": [0, 0],
        "ctf/tilt_A": [0, 0],
        "ctf/trefoil_A": [0, 0],
        "ctf/tetra_A": [0, 0, 0, 0],
        "ctf/anisomag": [0, 0, 0, 0],
        "ctf/bfactor": 0,
        "location/micrograph_shape": [4096, 4096],
        "alignments2D/psize_A": 2.1600000858306885,
        "alignments3D/psize_A": 2.1600000858306885
      }

I’m not sure where the “alignments_class_0/psize_A” value comes from, or why it’s identical to the Cs, but it seems to be governing the refinement. Subsequent NU-refinement with the same particles goes to higher resolution, with Nyquist frequency that better resembles to the pixel size:
image

I got similar results when I unbinned the particles to a 360 pixel box. Here the resolution and the particle metadata from heterogenous refinement looks like this:
image

      "passthrough": "particles",
      "num_items": 950822,
      "summary": {
        "alignments_class_0/psize_A": 3.0375001430511475,
        "alignments_class_1/psize_A": 3.0375001430511475,
        "blob/shape": [360, 360],
        "blob/psize_A": 1.0800000429153442,
        "blob/sign": -1,
        "ctf/type": "spline",
        "ctf/accel_kv": 300,
        "ctf/cs_mm": 2.700000047683716,
        "ctf/amp_contrast": 0.10000000149011612,
        "ctf/phase_shift_rad": 0,
        "ctf/scale": 1,
        "ctf/shift_A": [0, 0],
        "ctf/tilt_A": [0, 0],
        "ctf/trefoil_A": [0, 0],
        "ctf/tetra_A": [0, 0, 0, 0],
        "ctf/anisomag": [0, 0, 0, 0],
        "ctf/bfactor": 0,
        "location/micrograph_shape": [4096, 4096],
        "alignments2D/psize_A": 1.0800000429153442,
        "alignments3D/psize_A": 1.0800000429153442
      }

Any advice on where the different pixel sizes are coming from, and how to clean up my particle metadata, is much appreciated.

2

Hi,

You’ve probably kept the default Refinement box size (Voxels) value of 128 when building out your jobs. You can change this to match your input image dimensions.

Cheers,
Yang

1 Like
3

Of course you’re right :slight_smile: thanks