6. On the “Auto-sampling” tab, leave the default of “Initial
angular sampling: 1.8 degrees”, “Initial offset
range (pix): 3”, and “Initial offset step (pix):
0.75”. In general, one sets the sampling rates to somewhat
coarser values than the estimated accuracy as reported in the
“model_classes” table in the model.star file from the consensus refinement. For the spliceosome case, the values are
0.786
for the angular accuracy and 0.535 pixels for the translational accuracy, as given in the file “Example/consensus_half1_model.star”. (also see Note 7).
7. On the “Analyse” tab, leave the defaults. “Run flexibility
analyses? Yes” will perform the principal components analysis on the relative orientations of all bodies upon convergence.
“Number of eigenvector movies: 3” will select the first
three principal motions in the data, and write out ten maps
for each of them (see Note 8). “Select particles based on
eigenvalues? No” will switch off the option to select particle
subsets based on eigenvalues (see Note 9).
8. On the “Compute” tab, set “Use parallel disc I/O?” to
“Yes” if you have a fast, parallel storage system, or if you are
working on a local hard drive. Set this option to “No” if you are
using a less parallel, shared storage system like NFS.
9. On the “Compute” tab, set “Number of pooled particles:
30” if you are using GPUs. Otherwise, use a smaller value
like 3.
10. On the “Compute” tab, set “Skip padding? Yes”, as this will
save RAM and GPU memory by up to eight-fold. Only set this
option to “No” if your particle occupies such a large volume of
the box that artifacts are observed near the edges of the box in
runs where this option was set to “Yes”.
11. On the “Compute” tab, set “Pre-read all particles into
RAM? No”, as the example data set occupies more than 120 GB,
and each MPI worker would read the full data set into memory.
Only set this option to “Yes” if your data set is small enough to
fit into your computer’s RAM.
12. On the “Compute” tab, set “Copy particles to scratch
directory: /ssd”, provided your computer has a local SSD
scratch drive that is called “/ssd”, and reading from this drive
is significantly faster than reading from the storage system your
data resides on. If your data set is small enough to read into
RAM (previous step), then copying the data to a scratch drive is
not necessary.
13. On the “Compute” tab, set “Combine iterations through
disc? No”. Only set this option to “Yes” if the default option
of combining iterations through the network gives problems,
e.g., MPI-related errors at the end of an iteration.
Multi-body Refinement in RELION
151
angular sampling: 1.8 degrees”, “Initial offset
range (pix): 3”, and “Initial offset step (pix):
0.75”. In general, one sets the sampling rates to somewhat
coarser values than the estimated accuracy as reported in the
“model_classes” table in the model.star file from the consensus refinement. For the spliceosome case, the values are
0.786
for the angular accuracy and 0.535 pixels for the translational accuracy, as given in the file “Example/consensus_half1_model.star”. (also see Note 7).
7. On the “Analyse” tab, leave the defaults. “Run flexibility
analyses? Yes” will perform the principal components analysis on the relative orientations of all bodies upon convergence.
“Number of eigenvector movies: 3” will select the first
three principal motions in the data, and write out ten maps
for each of them (see Note 8). “Select particles based on
eigenvalues? No” will switch off the option to select particle
subsets based on eigenvalues (see Note 9).
8. On the “Compute” tab, set “Use parallel disc I/O?” to
“Yes” if you have a fast, parallel storage system, or if you are
working on a local hard drive. Set this option to “No” if you are
using a less parallel, shared storage system like NFS.
9. On the “Compute” tab, set “Number of pooled particles:
30” if you are using GPUs. Otherwise, use a smaller value
like 3.
10. On the “Compute” tab, set “Skip padding? Yes”, as this will
save RAM and GPU memory by up to eight-fold. Only set this
option to “No” if your particle occupies such a large volume of
the box that artifacts are observed near the edges of the box in
runs where this option was set to “Yes”.
11. On the “Compute” tab, set “Pre-read all particles into
RAM? No”, as the example data set occupies more than 120 GB,
and each MPI worker would read the full data set into memory.
Only set this option to “Yes” if your data set is small enough to
fit into your computer’s RAM.
12. On the “Compute” tab, set “Copy particles to scratch
directory: /ssd”, provided your computer has a local SSD
scratch drive that is called “/ssd”, and reading from this drive
is significantly faster than reading from the storage system your
data resides on. If your data set is small enough to read into
RAM (previous step), then copying the data to a scratch drive is
not necessary.
13. On the “Compute” tab, set “Combine iterations through
disc? No”. Only set this option to “Yes” if the default option
of combining iterations through the network gives problems,
e.g., MPI-related errors at the end of an iteration.
Multi-body Refinement in RELION
151