the situation in single particle cryo-EM [70]. Threshold constrained cross correlation [71] estimates the reliable set of Fourier components to compute cross correlation for each subtomogram which results in better alignment of particles to the
average and a slightly higher resolution. The commonly used empirical approach is
to separate the dataset into two parts, process them separately, and at each iteration
calculate to which resolution the two structures from the independent datasets
aligned to each other mach. Then this resolution is used as the resolution limit for
the following iteration. This scheme is implemented in multiple processing packages, including Dynamo [72], EMAN2 [73], and Relion [63]. Both approaches
described here deal with overfitting of noise.
The presence of dense cellular background or membranes inside the mask for
alignment may influence or dominate the alignment of particles to the template. The
use of sharp alignment masks with local cross-correlation was adjusted for particle
picking in single-particle EM in 2D [74] and for subtomogram averaging in 3D
[72]. Use of local cross-correlation minimizes the influence of the context on the
particle alignment and is beneficial for in situ structural analysis of protein complexes in crowded cellular environments or membrane proteins embedded into
membranes.
10.4.2 Limited Number of Particles
More particles increase the signal to noise ratio in single particle cryo-EM [10] with
a close-to-linear relation between the logarithm of particle number and the inverse
resolution [75]. In StA after reaching a certain number of particles and resolution
the other factors could come into the play and saturate the resolution [26]. The
currently reported structures at subnanometer resolution mostly included over
50,000 asymmetric subunits recorded on direct detectors. The particles that align
well to the average or have higher signal-to-noise ratio saturate the resolution faster,
as a result less particles are required for a given target resolution. Each of the
high-contrast particles recorded with the use of the Volta Phase Plate had resolutions from 35 to 50 Å while an average of 339 particles resulted in resolution of
31 Å, another conformation with 1367 particles had a resolution of 27 Å [32].
Subtomograms with large ribosome particles on the ER membranes recorded with
higher defocus have been shown to align very accurately, which helped in obtaining
subnanometer resolution with less particles [60]. However when the data is very
high quality and the molecule of interest is very structurally homogeneous large
number of particles may lead to atomic resolution [67]. In order to “pick” enough
particles from the tomograms manual, semi-automatic [76] and automatic [77]
methods have been suggested.
276
M. Kudryashev
average and a slightly higher resolution. The commonly used empirical approach is
to separate the dataset into two parts, process them separately, and at each iteration
calculate to which resolution the two structures from the independent datasets
aligned to each other mach. Then this resolution is used as the resolution limit for
the following iteration. This scheme is implemented in multiple processing packages, including Dynamo [72], EMAN2 [73], and Relion [63]. Both approaches
described here deal with overfitting of noise.
The presence of dense cellular background or membranes inside the mask for
alignment may influence or dominate the alignment of particles to the template. The
use of sharp alignment masks with local cross-correlation was adjusted for particle
picking in single-particle EM in 2D [74] and for subtomogram averaging in 3D
[72]. Use of local cross-correlation minimizes the influence of the context on the
particle alignment and is beneficial for in situ structural analysis of protein complexes in crowded cellular environments or membrane proteins embedded into
membranes.
10.4.2 Limited Number of Particles
More particles increase the signal to noise ratio in single particle cryo-EM [10] with
a close-to-linear relation between the logarithm of particle number and the inverse
resolution [75]. In StA after reaching a certain number of particles and resolution
the other factors could come into the play and saturate the resolution [26]. The
currently reported structures at subnanometer resolution mostly included over
50,000 asymmetric subunits recorded on direct detectors. The particles that align
well to the average or have higher signal-to-noise ratio saturate the resolution faster,
as a result less particles are required for a given target resolution. Each of the
high-contrast particles recorded with the use of the Volta Phase Plate had resolutions from 35 to 50 Å while an average of 339 particles resulted in resolution of
31 Å, another conformation with 1367 particles had a resolution of 27 Å [32].
Subtomograms with large ribosome particles on the ER membranes recorded with
higher defocus have been shown to align very accurately, which helped in obtaining
subnanometer resolution with less particles [60]. However when the data is very
high quality and the molecule of interest is very structurally homogeneous large
number of particles may lead to atomic resolution [67]. In order to “pick” enough
particles from the tomograms manual, semi-automatic [76] and automatic [77]
methods have been suggested.
276
M. Kudryashev