Preprocess images

RELION will work best if your data are

• Even-sized (Odd dimensions are not implemented)
• Clean from false particles (no images are discarded during refinement).
  • Xmipp implements an image sorting utility called xmipp_sort_by_statistics that is very handy in the cleaning of a data set.
• Unmasked (masking is performed internally)
• Non-interpolated (prevent any prior rotations/translations: use the originally scanned pixel values)
  • If downscaling is necessary because of memory issues: use a window-operation in Fourier-space, not a convolution in real-space (e.g. with rectangle/B-spline).
  • Xmipp implements the Fourier-space downscaling in the xmipp_scale program with the -fourier option.
• Uncorrected for CTF (this is done internally)
  • If your data have previously been phase-flipped, that's OK: just tell RELION about it
  • Actually, if you are not planning to correct for CTFs inside RELION (e.g. for negative stain data), phase-flipping is recommended.
  • If your data have previously been pre-Wiener filtered or pre-multiplied by their CTF, that's a bad thing to do: go back to the original data.
• Normalised Make sure the average density in the background area is (approximately) zero!!. The exact procedure probably does not matter too much, as errors in the normalisation are corrected internally, but large positive or negative values average background densities .

And then, just like with any other refinement program, you might save yourself lots of trouble if your data have:

• high signal-to-noise ratios (get the best possible structure by taking great care in sample preparation and data collection)

FAQs

• Do I need to downsize my images?
  • RELION will estimate the resolution of your model and downsize the images internally. Therefore, using downsized images will NOT be much faster than using large images. The only reason why one would want to downsize images if they are very big (e.g. 800x800 pixels) and the reconstruction would not fit into memory.