Scripts from av3/tom are in MATLAB (ending:.m) and PyTom procedures in
python (.py). The main steps are (I) preparing files for template matching, (II)
running template matching on your local computing cluster, (III) extracting peaks
from the cross-correlation volumes and (IV) selecting peaks for subtomogram
reconstruction.
(I) Preparation of files for template matching:
– A template structure can be generated either from EM densities deposited
in the EMDB or by simulating EM density based on an atomic structure
deposited in the PDB (tom_pdb2em.m). In order to recapitulate the
imaging conditions used for acquisition of the tomography data, the
density should be convoluted with a simulated CTF (tom_create_ctf.m).
To reduce template bias, the template structure should be low-pass filtered to 5 nm resolution (tom_bandpass.m), before being scaled to the
appropriate voxel size (tom_rescale.m). The resulting template structure
can also be mirrored (tom_mirror.m) to probe specificity of detections.
– In addition to the template structure, a mask has to be prepared (tom_spheremask.m). The mask must have the same dimensions as the template
structure and should encompass the template tightly without cutting off
density. The edges of the mask should be smoothened to reduce artifacts
from masking.
– A job file has to be prepared, in which the paths for the tomogram to be
queried, the template structure and the mask are specified. Furthermore,
the missing wedge of the tomogram, the type of scoring function to be
used, and the angular sampling have to be specified. Templates for
generating the job file can be downloaded from www.pytom.org.
(II) Run template matching on your local computing cluster (localization.py).
Template matching outputs two volumes in the starting directory:
– scores.em, containing the constrained cross-correlation values for the
best-matching orientation for each voxel.
– angles.em, containing the orientation of the template structure that yielded the best cross-correlation for each voxel.
(III) Extract peaks from the correlation volume into an av3/TOM motiflist
(av3_createmotl.m) or PyTom XML particle list. Particular attention should
be paid to the radius of the sphere used to mask out the correlation volume
around the respective peak. This radius should be chosen as big as possible,
to avoid picking the same copy of your complex of interest several times, but
without risking to mask out peaks of neighboring copies. Typically, the
number of peaks being extracted should be considerably higher than the
anticipated number of instances of your complex of interest.
(IV) For selection of peaks for subtomogram reconstruction, three approaches are
commonly used. Fast ways are to select peaks according to (A) the distribution
of cross-correlation coefficients or (B) a comparison of cross-correlation
252
S. Pfeffer and F. Förster
python (.py). The main steps are (I) preparing files for template matching, (II)
running template matching on your local computing cluster, (III) extracting peaks
from the cross-correlation volumes and (IV) selecting peaks for subtomogram
reconstruction.
(I) Preparation of files for template matching:
– A template structure can be generated either from EM densities deposited
in the EMDB or by simulating EM density based on an atomic structure
deposited in the PDB (tom_pdb2em.m). In order to recapitulate the
imaging conditions used for acquisition of the tomography data, the
density should be convoluted with a simulated CTF (tom_create_ctf.m).
To reduce template bias, the template structure should be low-pass filtered to 5 nm resolution (tom_bandpass.m), before being scaled to the
appropriate voxel size (tom_rescale.m). The resulting template structure
can also be mirrored (tom_mirror.m) to probe specificity of detections.
– In addition to the template structure, a mask has to be prepared (tom_spheremask.m). The mask must have the same dimensions as the template
structure and should encompass the template tightly without cutting off
density. The edges of the mask should be smoothened to reduce artifacts
from masking.
– A job file has to be prepared, in which the paths for the tomogram to be
queried, the template structure and the mask are specified. Furthermore,
the missing wedge of the tomogram, the type of scoring function to be
used, and the angular sampling have to be specified. Templates for
generating the job file can be downloaded from www.pytom.org.
(II) Run template matching on your local computing cluster (localization.py).
Template matching outputs two volumes in the starting directory:
– scores.em, containing the constrained cross-correlation values for the
best-matching orientation for each voxel.
– angles.em, containing the orientation of the template structure that yielded the best cross-correlation for each voxel.
(III) Extract peaks from the correlation volume into an av3/TOM motiflist
(av3_createmotl.m) or PyTom XML particle list. Particular attention should
be paid to the radius of the sphere used to mask out the correlation volume
around the respective peak. This radius should be chosen as big as possible,
to avoid picking the same copy of your complex of interest several times, but
without risking to mask out peaks of neighboring copies. Typically, the
number of peaks being extracted should be considerably higher than the
anticipated number of instances of your complex of interest.
(IV) For selection of peaks for subtomogram reconstruction, three approaches are
commonly used. Fast ways are to select peaks according to (A) the distribution
of cross-correlation coefficients or (B) a comparison of cross-correlation
252
S. Pfeffer and F. Förster