cross-correlation coefficients and (III) using this matrix for CPCA and kmeans
clustering.
(I) Preparation of files for subtomogram classification:
– Subtomograms have to be pre-aligned to a common reference system to
obtain an aligned PyTom XML particle list.
– A mask has to be prepared (tom_spheremask.m) that focuses on the
feature of interest, but it should not be chosen too small.
– A job file has to be prepared that specifies the paths for the aligned
PyTom XML particle list and the mask, as well as the lowpass filter for
used for classification. The lowpass filter should be set according to the
size of the feature to be classified for. Since single subtomograms have a
low SNR, higher frequencies should be generally excluded, but the feature of interest must not be compromised.
(II) The pairwise matrix of constrained cross-correlation coefficients (output file:
correlation_matrix.csv) is computed using PyTom on your local computing
cluster (calculate_correlation_matrix.py). Computational time will significantly increase with the number and size of subtomograms.
(III) Principal component analysis and k-means clustering are performed in
av3/TOM (pcacov.m, kmeans.m from the MATLAB statistics toolbox). The
number of eigenvectors for CPCA, the number of classes for k-means
clustering and the subtomogram path have to be specified. Typically, 5
eigenvectors are sufficient to represent the main features of the data and
suppress the noise. The number of classes should be chosen as high as
possible in order to be able to pick up also small populations of structurally
distinct particles. For high quality data, 200 subtomograms per class usually
yield sufficient SNR for subsequent visual assessment or further clustering
using hierarchical classification, treating the class averages like subtomograms. Once sufficiently pure subsets of particles have been obtained, they
can be averaged separately to increase the SNR for each distinct population.
References
1. R. Henderson, The potential and limitations of neutrons, electrons and X-rays for atomic
resolution microscopy of unstained biological molecules. Q. Rev. Biophys. 28(2), 171–193
(1995)
2. K. Grunewald, P. Desai, D.C. Winkler, J.B. Heymann, D.M. Belnap, W. Baumeister, A.C.
Steven, Three-dimensional structure of herpes simplex virus from cryo-electron tomography.
Science 302(5649), 1396–1398 (2003). doi:10.1126/science.1090284
3. S. Asano, Y. Fukuda, F. Beck, A. Aufderheide, F. Forster, R. Danev, W. Baumeister,
Proteasomes. A molecular census of 26S proteasomes in intact neurons. Science 347(6220),
439–442 (2015). doi:10.1126/science.1261197
9 Structural Biology in Situ Using Cryo-Electron …
255
clustering.
(I) Preparation of files for subtomogram classification:
– Subtomograms have to be pre-aligned to a common reference system to
obtain an aligned PyTom XML particle list.
– A mask has to be prepared (tom_spheremask.m) that focuses on the
feature of interest, but it should not be chosen too small.
– A job file has to be prepared that specifies the paths for the aligned
PyTom XML particle list and the mask, as well as the lowpass filter for
used for classification. The lowpass filter should be set according to the
size of the feature to be classified for. Since single subtomograms have a
low SNR, higher frequencies should be generally excluded, but the feature of interest must not be compromised.
(II) The pairwise matrix of constrained cross-correlation coefficients (output file:
correlation_matrix.csv) is computed using PyTom on your local computing
cluster (calculate_correlation_matrix.py). Computational time will significantly increase with the number and size of subtomograms.
(III) Principal component analysis and k-means clustering are performed in
av3/TOM (pcacov.m, kmeans.m from the MATLAB statistics toolbox). The
number of eigenvectors for CPCA, the number of classes for k-means
clustering and the subtomogram path have to be specified. Typically, 5
eigenvectors are sufficient to represent the main features of the data and
suppress the noise. The number of classes should be chosen as high as
possible in order to be able to pick up also small populations of structurally
distinct particles. For high quality data, 200 subtomograms per class usually
yield sufficient SNR for subsequent visual assessment or further clustering
using hierarchical classification, treating the class averages like subtomograms. Once sufficiently pure subsets of particles have been obtained, they
can be averaged separately to increase the SNR for each distinct population.
References
1. R. Henderson, The potential and limitations of neutrons, electrons and X-rays for atomic
resolution microscopy of unstained biological molecules. Q. Rev. Biophys. 28(2), 171–193
(1995)
2. K. Grunewald, P. Desai, D.C. Winkler, J.B. Heymann, D.M. Belnap, W. Baumeister, A.C.
Steven, Three-dimensional structure of herpes simplex virus from cryo-electron tomography.
Science 302(5649), 1396–1398 (2003). doi:10.1126/science.1090284
3. S. Asano, Y. Fukuda, F. Beck, A. Aufderheide, F. Forster, R. Danev, W. Baumeister,
Proteasomes. A molecular census of 26S proteasomes in intact neurons. Science 347(6220),
439–442 (2015). doi:10.1126/science.1261197
9 Structural Biology in Situ Using Cryo-Electron …
255