algorithms, but also on the angular distribution of the particles. Hence, in order to
get the best resolution reconstruction, it is necessary for the particle (and thus its
projections) to be distributed well in the Euler sphere [41]. By re-projecting the 3D
reconstruction at the Euler angles of the class averages, we can assess the reliability
of the 3D reconstruction. For a consistent reliable reconstruction, the re-projected
image and the actual class average must match.
3 Resolution, Model Building, and Validation
3.1 Resolution
Resolution estimation of the EM maps is still subjective, with differences among
various groups still not settled [76]. Resolution of 3D EM map is calculated from a
plot of Fourier shell correlation (FSC) [77] as a function of spatial frequency (the
resolution estimation of 3D reconstructions in Fig. 4 is shown in Fig. 5). FSC is the
cross-correlation (CC) calculated between two 3D reconstruction maps, where each
map is calculated from half the data images. The resolution that is reported in
publication essentially as a single number is the value of maximum spatial frequency up to which the EM map is reliable. The identification of resolution is
subjective as it is arbitrary what one considers as reliable. The procedure for resolution assessment is described in detail by Penczek [76]. There are several suggestions for identifying the cutoff: (i) the 3-sigma criteria where the spectral SNR
(SSNR) = 0 in which case FSC = 0; (ii) point at which power of signal is equal to
the power of noise, i.e., SSNR = 1 or FSC = 0.33; (iii) the classic midpoint of FSC
curve, i.e., FSC = 0.5 [78] where SSNR = 2, which means signal dominates noise;
and finally (iv) point where FSC = 0.143, derived by Rosenthal and Henderson
[79] in comparison with X-ray crystallography. Hence, which cutoff is chosen is a
matter of present-day debate. Recently, in order to reduce further any possible
reference bias, “gold-standard FSC” was suggested with FSC calculated between
two completely independent refinements and 3D reconstruction [80].
There are other computational ways to improve the resolution nominally without
improving the image alignment, e.g., masking/threshold flattening. In any case, the
resolution estimations have their own limitations and hence reported EM resolution
should be treated as only broad guideline rather than a definitive number and cannot
be used as validation. Nonetheless, it is an important parameter to be reported with
each EM map deposition at the EMDB. Resolution anisotropy is common in
cryo-EM structures, and it is a common practice to document it as color ramping
from low to high resolution on the cryo-EM 3D reconstruction map using programs
ResMap [81] and blocres [82]. The results can be visualized independently or with
chimera (e.g., blocres with Local FSC plug-in for chimera).
With the booming medium- and high-resolution cryo-EM 3D structures, it is
necessary to have consistency between crystallography and cryo-EM terms
Single-Particle cryo-EM as a Pipeline for Obtaining Atomic …
389
get the best resolution reconstruction, it is necessary for the particle (and thus its
projections) to be distributed well in the Euler sphere [41]. By re-projecting the 3D
reconstruction at the Euler angles of the class averages, we can assess the reliability
of the 3D reconstruction. For a consistent reliable reconstruction, the re-projected
image and the actual class average must match.
3 Resolution, Model Building, and Validation
3.1 Resolution
Resolution estimation of the EM maps is still subjective, with differences among
various groups still not settled [76]. Resolution of 3D EM map is calculated from a
plot of Fourier shell correlation (FSC) [77] as a function of spatial frequency (the
resolution estimation of 3D reconstructions in Fig. 4 is shown in Fig. 5). FSC is the
cross-correlation (CC) calculated between two 3D reconstruction maps, where each
map is calculated from half the data images. The resolution that is reported in
publication essentially as a single number is the value of maximum spatial frequency up to which the EM map is reliable. The identification of resolution is
subjective as it is arbitrary what one considers as reliable. The procedure for resolution assessment is described in detail by Penczek [76]. There are several suggestions for identifying the cutoff: (i) the 3-sigma criteria where the spectral SNR
(SSNR) = 0 in which case FSC = 0; (ii) point at which power of signal is equal to
the power of noise, i.e., SSNR = 1 or FSC = 0.33; (iii) the classic midpoint of FSC
curve, i.e., FSC = 0.5 [78] where SSNR = 2, which means signal dominates noise;
and finally (iv) point where FSC = 0.143, derived by Rosenthal and Henderson
[79] in comparison with X-ray crystallography. Hence, which cutoff is chosen is a
matter of present-day debate. Recently, in order to reduce further any possible
reference bias, “gold-standard FSC” was suggested with FSC calculated between
two completely independent refinements and 3D reconstruction [80].
There are other computational ways to improve the resolution nominally without
improving the image alignment, e.g., masking/threshold flattening. In any case, the
resolution estimations have their own limitations and hence reported EM resolution
should be treated as only broad guideline rather than a definitive number and cannot
be used as validation. Nonetheless, it is an important parameter to be reported with
each EM map deposition at the EMDB. Resolution anisotropy is common in
cryo-EM structures, and it is a common practice to document it as color ramping
from low to high resolution on the cryo-EM 3D reconstruction map using programs
ResMap [81] and blocres [82]. The results can be visualized independently or with
chimera (e.g., blocres with Local FSC plug-in for chimera).
With the booming medium- and high-resolution cryo-EM 3D structures, it is
necessary to have consistency between crystallography and cryo-EM terms
Single-Particle cryo-EM as a Pipeline for Obtaining Atomic …
389