Processing EIGER images by DIALS

This has nothing to do with SACLA but may be useful for serial crystallography at synchrotrons.

We use dials.stills_process to process still images. However, before starting batch processing, we first do it manually to find parameters.

Importing images

As with rotation dataset, we use dials.import. You can override wrong information in the master h5 as follows.

dials.import stills_master.h5 geometry.beam.wavelength=1.00 \
   panel.origin=-33.73,-39.96,-78 panel.fast_axis=1,0,0 panel.slow_axis=0,1,0

If you have a CrystFEL geometry file, you can calculate the panel origin by multiplying corner_x and corner_y by the pixel size (the inverse of res). For example, in the case above, the relevant sections in the geometry file are:

res        = 13333.3
0/corner_x = -449.692089
0/corner_y = -532.831936

You can also use the circle fitting feature in iMOSFLM to find the beam center. Note that you have to swap X and Y from iMOSFLM.

Generating the pixel mask

DIALS respects the pixel mask in the master h5. If you want to add more masks (i.e. beam stop, rings, etc), you can use dials.generate_mask. untrusted.rectangle specifies coordinates in the x0,x1,y1,y2 order.

dials.generate_mask datablock.json resolution_range=2.85,2.91 resolution_range=2.45,2.5 \
   untrusted.rectangle=448,462,255,516

This creates mask.pickle. Repeat dials.import with mask=mask.pickle.

dials.import stills_master.h5 geometry.beam.wavelength=1.00 \
   panel.origin=-33.73,-39.96,-78 panel.fast_axis=1,0,0 panel.slow_axis=0,1,0 \
   mask=mask.pickle

This mask does not inherit the pixel mask in the master H5, but does apply the trusted range mask by default (use_trusted_range=True). The mask in the master h5 is merged during the processing (see https://github.com/dials/dials/issues/236).

Spot finding

dials.find_spots datablock.json

Check spot finding results:

dials.image_viewer datablock.json strong.pickle

When the result is not satisfactory, we typically tweak gain, min_spot_size and global_threshold.

Indexing

dials.index datablock.json strong.pickle space_group=P43212  unit_cell=79.4,79.4,37.8 \
 d_min_start=2.0 indexing.method=real_space_grid_search ewald_proximal_volume_max=0.05 \
 scan_range=1,2

We put scan_range=1,2 to index only the first image in the container. I don't know why we sometime need to increase ewald_proximal_volume_max.

If real_space_grid_search rejects too many spots (far fewer spots in refinement than observed), try setting both index_assignment.method=local and optimise_initial_basis_vectors=True. It is also worth trying indexing.method=fft1d.

When you index rotation datasets, the refinement is repeated with higher and higher resolution limits, starting from d_min_start. In contrast, the still indexer does not repeat the refinement. Thus, you should specify reasonably high resolution in the parameter. It also limits the spots real_space_grid_search examines.

If you see RMSD_Z in the output, your images are not treated as stills, but as rotation images. Add stills.indexer = stills to the arguments.

Batch processing

When you are happy with parameters, create index.phil for dials.stills_process.

# set to the number of cores in your computer
mp.nproc = 14

# get more info
verbosity = 3

geometry {
    detector {
        panel {
            origin=-33.73,-39.96,-78 
            fast_axis=1,0,0 
            slow_axis=0,1,0
        }
    }
    beam.wavelength=1.00
}

# we can specify `datablock.json` created above to avoid repetition.
# HOWEVER, you still have to specify wavelength and masks as these are not read from the file.
# input.reference_geometry = datablock.json

indexing {
    # force stills mode
    stills.indexer = stills

    known_symmetry {
        space_group = P43212
        unit_cell = 78.9 78.9 38.1 90 90 90
    }
    method = fft1d

    # sometimes real_space_grid_search works better
    # method = real_space_grid_search

    # needs to be sufficiently high
    refinement_protocol.d_min_start = 2.0

    # dramatically increases accepted spots in real_space_grid_search
    # index_assignment.method = local
    # optimise_initial_basis_vectors = True

    # don't know why this is necessary in some cases
    stills.ewald_proximal_volume_max = 0.05
}

spotfinder {
    filter.min_spot_size = 2

    # don't forget to specify it in the "integration" section as well!
    lookup.mask = mask.pickle

}

refinement {
    parameterisation {
        # not sure which is better. The default is False
        # spherical_relp_model = True

        beam.fix = all
        detector.fix = all
        auto_reduction {
            action = fix
            min_nref_per_parameter = 1
        }
        crystal {
            unit_cell {
                restraints {
                    tie_to_target {
                        values = 78.9,78.9,38.1,90,90,90
                        sigmas = 1,1,1,0,0,0
                    }
                }
            }
        }
    }
}

integration {
    lookup.mask = mask.pickle

    integrator = stills
    profile.fitting = False
    background {
        algorithm = simple
        simple {
            model.algorithm = linear2d
            outlier.algorithm = simple
        }
    }
}

profile {
    gaussian_rs {
        min_spots.overall = 0
    }   
}

Then execute:

dials.stills_process index.phil stills_master.h5