of the complete unbinned tomographic volume. Scaling factors originating
from size reduction of the tomographic volume and offsets originating from
cutting the volume have to be considered. Subtomograms should be large
enough to easily include the macromolecules of interest, typically twice their
size.
(II) Three distinct approaches for iterative alignment of subtomograms are
implemented in PyTom. For all three alignment approaches a reference
structure for the first iteration, a mask and a PyTom XML particle list must be
prepared:
– A reference structure can be either generated from an external structure as
described above (template matching, preparation of files) or it can be
obtained by averaging the reconstructed subtomograms, using the orientations determined by template matching (av3_average_exact.m).
– A mask can be prepared as described above (template matching, preparation of files).
– A PyTom XML particle list can be prepared from a TOM/av3 motiflist in
PyTom (fromMOTL.py and toXMLFile.py).
In PyTom, subtomograms can be aligned using coupled translational and
restricted rotational search in real space following either a ‘conventional’
(ExMaxAlignment.py) or a ‘gold standard’ alignment approach (GLocalJob.py).
Alternatively, subtomograms can be aligned using alternating translational and
global rotational search in Fourier space using spherical harmonics
(FRMAlignment.py). For each alignment approach, specific job files have to be
prepared. In these job files, you have to specify the output folder, the number of
alignment iterations, the paths for reference structure and mask, the voxel size, the
bandpass filter and the angular increment of rotational search for the first iteration
(not in FRM). Running ‘conventional’ alignment, the PyTom XML particle list
generated for the selected peaks has to be inserted into the job file at the indicated
position. Running ‘gold standard’ or FRM alignment, a path can specify the
PyTom XML particle list. Templates for generating the job files, as well as further
instructions for choosing appropriate alignment parameters can be downloaded
from www.pytom.org.
9.8.3 Classification of Subtomograms Using CPCA
and Kmeans Clustering
This section introduces how to sort out compositional and conformational heterogeneity in a set of pre-aligned subtomograms using constrained principal component analysis and k-means clustering in av3/PyTom. The main steps are
(I) preparing files for classification, (II) computing a pairwise matrix of constrained
254
S. Pfeffer and F. Förster
from size reduction of the tomographic volume and offsets originating from
cutting the volume have to be considered. Subtomograms should be large
enough to easily include the macromolecules of interest, typically twice their
size.
(II) Three distinct approaches for iterative alignment of subtomograms are
implemented in PyTom. For all three alignment approaches a reference
structure for the first iteration, a mask and a PyTom XML particle list must be
prepared:
– A reference structure can be either generated from an external structure as
described above (template matching, preparation of files) or it can be
obtained by averaging the reconstructed subtomograms, using the orientations determined by template matching (av3_average_exact.m).
– A mask can be prepared as described above (template matching, preparation of files).
– A PyTom XML particle list can be prepared from a TOM/av3 motiflist in
PyTom (fromMOTL.py and toXMLFile.py).
In PyTom, subtomograms can be aligned using coupled translational and
restricted rotational search in real space following either a ‘conventional’
(ExMaxAlignment.py) or a ‘gold standard’ alignment approach (GLocalJob.py).
Alternatively, subtomograms can be aligned using alternating translational and
global rotational search in Fourier space using spherical harmonics
(FRMAlignment.py). For each alignment approach, specific job files have to be
prepared. In these job files, you have to specify the output folder, the number of
alignment iterations, the paths for reference structure and mask, the voxel size, the
bandpass filter and the angular increment of rotational search for the first iteration
(not in FRM). Running ‘conventional’ alignment, the PyTom XML particle list
generated for the selected peaks has to be inserted into the job file at the indicated
position. Running ‘gold standard’ or FRM alignment, a path can specify the
PyTom XML particle list. Templates for generating the job files, as well as further
instructions for choosing appropriate alignment parameters can be downloaded
from www.pytom.org.
9.8.3 Classification of Subtomograms Using CPCA
and Kmeans Clustering
This section introduces how to sort out compositional and conformational heterogeneity in a set of pre-aligned subtomograms using constrained principal component analysis and k-means clustering in av3/PyTom. The main steps are
(I) preparing files for classification, (II) computing a pairwise matrix of constrained
254
S. Pfeffer and F. Förster