[65] in the year 1998 and other program suites followed. The clustering of similar
particle images was first introduced by van Heel and Frank [68] in the year 1981
using multivariate statistical analysis. Clustering in the currently available programs
uses one of the following methods: multivariate statistical analysis (MSA)/principle
component analysis (PCA), hierarchical clustering, k-means clustering, and the
maximum-likelihood methods [41] or by recently proposed empirical Bayesian
approach [69]. Currently, the EMAN2 [49] and RELION-3 [64] are among the
popular program that do reference-free 2D class-averaging (references are generated from within the data set) and 3D reconstruction. EMAN2 uses iterative
MSA-based reference-free 2D classification. The latest one, the RELION, uses
empirical Bayesian likelihood approach for 2D classification [55].
Next step is to get the 3D reconstruction from selected good class averages.
High-resolution 3D reconstructions require an initial 3D model that can be iteratively refined to obtain the best possible resolution for the data set. The first starting
Fig. 4 Asymmetric (C1) 3D reconstructions of the three classes (structures). Each class (class 1
(a); class 2 (b); and class 3 (c)) is shown as a top view (top ring only), a side view, a central section
through the side view, and a bottom view (bottom ring only). The fitted GroEL crystal structure is
shown in green. The additional density in the upper rings of (a) and (b) is attributable to bound
non-native RuBisCO substrate. All maps were contoured at the 1r level without filtering.
Figure generated with Chimera [85]. Figure produced by the author in https://doi.org/10.1016/j.
ijbiomac.2018.06.120 [74] and reproduced here under Creative Commons Attribution License (CC
BY)
Single-Particle cryo-EM as a Pipeline for Obtaining Atomic …
387
particle images was first introduced by van Heel and Frank [68] in the year 1981
using multivariate statistical analysis. Clustering in the currently available programs
uses one of the following methods: multivariate statistical analysis (MSA)/principle
component analysis (PCA), hierarchical clustering, k-means clustering, and the
maximum-likelihood methods [41] or by recently proposed empirical Bayesian
approach [69]. Currently, the EMAN2 [49] and RELION-3 [64] are among the
popular program that do reference-free 2D class-averaging (references are generated from within the data set) and 3D reconstruction. EMAN2 uses iterative
MSA-based reference-free 2D classification. The latest one, the RELION, uses
empirical Bayesian likelihood approach for 2D classification [55].
Next step is to get the 3D reconstruction from selected good class averages.
High-resolution 3D reconstructions require an initial 3D model that can be iteratively refined to obtain the best possible resolution for the data set. The first starting
Fig. 4 Asymmetric (C1) 3D reconstructions of the three classes (structures). Each class (class 1
(a); class 2 (b); and class 3 (c)) is shown as a top view (top ring only), a side view, a central section
through the side view, and a bottom view (bottom ring only). The fitted GroEL crystal structure is
shown in green. The additional density in the upper rings of (a) and (b) is attributable to bound
non-native RuBisCO substrate. All maps were contoured at the 1r level without filtering.
Figure generated with Chimera [85]. Figure produced by the author in https://doi.org/10.1016/j.
ijbiomac.2018.06.120 [74] and reproduced here under Creative Commons Attribution License (CC
BY)
Single-Particle cryo-EM as a Pipeline for Obtaining Atomic …
387