solvent mask to half-maps in which the phases of the
high-resolution Fourier components have been randomized
provides a measure of the second effect. To this purpose, the
post-processing program randomizes the phases of the Fourier
components with resolution beyond the one where the green
FSC curve drops below 0.8. Then, it applies the solvent mask
to both phase-randomized maps and calculates the red FSC
curve between the two masked phase-randomized maps. Any
non-zero FSC values in the red curve beyond the resolution at
which the phases were randomized are due to the convolution
effect of the solvent mask. The black FSC curve applies a
correction for this effect to the blue curve. The spatial frequency at which this curve drops below 0.143 is the estimated
resolution of the combined map. However, the correction can
lead to artifacts, most notably under-estimation of resolution,
if the red curve does not drop to values close to zero near the
final estimated resolution of the map. In particular sharper
masks that cut through density in the half-maps lead to larger
FSC values in the red curves. Therefore, such masks should be
avoided.
17. During multi-body refinement, phase randomization is applied
for each body at every iteration to estimate its resolution. It is
possible that during the earlier iterations, when some bodies
have more spread-out, blurred densities, that the body masks
are too tight and the resolution for these bodies is underestimated due to the artifacts mentioned in Note 16. Sometimes these problems disappear by themselves as the body
reconstruction improves. In other cases, it may be necessary
to use wider and/or softer-edged masks. The “Post-processing” job-type described in Subheading 3.3, can also be
performed on the unfiltered half-maps that are written out
during the multi-body refinement as: “Multibody/job001/
run_it0??_half?_body00?_unfil.mrc”. However, during RELION refinements maps are not necessarily calculated
to the Nyquist frequency to save computational resources. This
may lead to zero values for all higher spatial frequency components in both half-maps, which in turn may result in artificially
high FSC values at these frequencies. Therefore, the resulting
FSC curves should only be evaluated up to the spatial frequency that was included in that iteration of the refinement,
which may be found in the corresponding model.star files.
18. As the quality or the local resolution of the density may vary
within each body, it may be useful to calculate multiple postprocessed maps for atomic model building. To this end, use the
options to provide a user-defined B-factor or a user-defined
spatial frequency for a low-pass filter on the “Sharpen” and
“Filter” tabs of the “Post-processing” job-type. In
Multi-body Refinement in RELION
157
high-resolution Fourier components have been randomized
provides a measure of the second effect. To this purpose, the
post-processing program randomizes the phases of the Fourier
components with resolution beyond the one where the green
FSC curve drops below 0.8. Then, it applies the solvent mask
to both phase-randomized maps and calculates the red FSC
curve between the two masked phase-randomized maps. Any
non-zero FSC values in the red curve beyond the resolution at
which the phases were randomized are due to the convolution
effect of the solvent mask. The black FSC curve applies a
correction for this effect to the blue curve. The spatial frequency at which this curve drops below 0.143 is the estimated
resolution of the combined map. However, the correction can
lead to artifacts, most notably under-estimation of resolution,
if the red curve does not drop to values close to zero near the
final estimated resolution of the map. In particular sharper
masks that cut through density in the half-maps lead to larger
FSC values in the red curves. Therefore, such masks should be
avoided.
17. During multi-body refinement, phase randomization is applied
for each body at every iteration to estimate its resolution. It is
possible that during the earlier iterations, when some bodies
have more spread-out, blurred densities, that the body masks
are too tight and the resolution for these bodies is underestimated due to the artifacts mentioned in Note 16. Sometimes these problems disappear by themselves as the body
reconstruction improves. In other cases, it may be necessary
to use wider and/or softer-edged masks. The “Post-processing” job-type described in Subheading 3.3, can also be
performed on the unfiltered half-maps that are written out
during the multi-body refinement as: “Multibody/job001/
run_it0??_half?_body00?_unfil.mrc”. However, during RELION refinements maps are not necessarily calculated
to the Nyquist frequency to save computational resources. This
may lead to zero values for all higher spatial frequency components in both half-maps, which in turn may result in artificially
high FSC values at these frequencies. Therefore, the resulting
FSC curves should only be evaluated up to the spatial frequency that was included in that iteration of the refinement,
which may be found in the corresponding model.star files.
18. As the quality or the local resolution of the density may vary
within each body, it may be useful to calculate multiple postprocessed maps for atomic model building. To this end, use the
options to provide a user-defined B-factor or a user-defined
spatial frequency for a low-pass filter on the “Sharpen” and
“Filter” tabs of the “Post-processing” job-type. In
Multi-body Refinement in RELION
157