of softwares are available for manual and automatic picking of particles and subsequent image processing. Examples of such programs are FindEM [48] (only for
automated particle picking), EMAN (e2boxer.py) [49, 50], IMAGIC [51], Ximdisp
[52] (only for interactive display, analyses, and particle picking; now a part of
CCP-EM package [53, 54]), Xmipp [30], RELION-autopick [55], cryoSPARC
[56], APPLE Picker [57] (completely automatic particle picking, a part of ASPIRE
Suite [58]), gEMpicker [59] (only for template-based particle picking),
SIGNATURE [60] (only for particle picking and data analysis), etc. Most of the
auto-picking software employ initial manual picking routine (except APPLE
picker), where a couple of thousands of particles are manually picked from a subset
of available micrographs and use the best class averages generated from them
(having as many different representative orientations) as templates to auto-pick
particles from rest of the micrographs. This is the preferred method. Alternatively,
the auto-picking programs can use low-pass-filtered EM maps as templates for
particle picking (less preferred, but useful in protein drug complex where you have
the apo-protein structure already). Using maps from PDB (Protein Data Bank)
coordinates, as reference model is not preferred at this stage in order to avoid
“Einstein-from-noise” effect [61], i.e., to avoid any 2D model bias. CTF corrections
can also be performed on picked particle images as compared to whole micrographs
in some software, e.g., EMAN [44].
After particle picking, the next stage is to get the 3D reconstruction of the
biological macromolecules using the different but identifiable 2D projections of
particles. The first 3D reconstruction from a 2D projection was carried out on
negative stained tail of bacteriophage T4 by De Rosier and Klug [62]. However, the
2D projections of particle images cutout from the motion-corrected micrographs
have still low signal-to-noise ratio (SNR) due to low electron dose data collection as
described in data collection section. Hence, in order to improve the SNR of the
particles, many identical looking particle images are aligned and summed (clustering) thus effectively increasing the SNR and dose without increasing the damage
[62]. There are three main advantages of reference-free (unsupervised) 2D classification: (i) to select few 2D classes from which we can make starting 3D map,
which can be projected as references for refinement. (ii) We can identify the fraction
of bad classes (which may contain artifacts, invalid particles, or simply empty), and
thus, those images with anomalies can be deleted from the data set in the beginning
itself. (iii) It also helps in identifying the conformational and compositional variability in the data set [50]. Two-dimensional (2D) and 3D classifications are carried
out by using various statistical analysis software suite IMAGIC [51], Spider [63],
EMAN [44], RELION-3 [64], FREALIGN [65], Appion [39], cryoSPARC [56],
ASPIRE Suite [58], Xmipp [30], SPHIRE (sphire.mpg.de), etc., or a combination of
more than one of these suites. Several of these software packages are integrated into
one processing framework, for example, as in Scipion [66]. An exhaustive list of
EM software programs is available at EMDataBank (EMDB, http://www.
emdatabank.org/emsoftware.html).
Spider [63, 67] and IMAGIC [51] were among the first programs to be developed for single-particle reconstruction in the year 1996 followed by FREALIGN
Single-Particle cryo-EM as a Pipeline for Obtaining Atomic …
385
automated particle picking), EMAN (e2boxer.py) [49, 50], IMAGIC [51], Ximdisp
[52] (only for interactive display, analyses, and particle picking; now a part of
CCP-EM package [53, 54]), Xmipp [30], RELION-autopick [55], cryoSPARC
[56], APPLE Picker [57] (completely automatic particle picking, a part of ASPIRE
Suite [58]), gEMpicker [59] (only for template-based particle picking),
SIGNATURE [60] (only for particle picking and data analysis), etc. Most of the
auto-picking software employ initial manual picking routine (except APPLE
picker), where a couple of thousands of particles are manually picked from a subset
of available micrographs and use the best class averages generated from them
(having as many different representative orientations) as templates to auto-pick
particles from rest of the micrographs. This is the preferred method. Alternatively,
the auto-picking programs can use low-pass-filtered EM maps as templates for
particle picking (less preferred, but useful in protein drug complex where you have
the apo-protein structure already). Using maps from PDB (Protein Data Bank)
coordinates, as reference model is not preferred at this stage in order to avoid
“Einstein-from-noise” effect [61], i.e., to avoid any 2D model bias. CTF corrections
can also be performed on picked particle images as compared to whole micrographs
in some software, e.g., EMAN [44].
After particle picking, the next stage is to get the 3D reconstruction of the
biological macromolecules using the different but identifiable 2D projections of
particles. The first 3D reconstruction from a 2D projection was carried out on
negative stained tail of bacteriophage T4 by De Rosier and Klug [62]. However, the
2D projections of particle images cutout from the motion-corrected micrographs
have still low signal-to-noise ratio (SNR) due to low electron dose data collection as
described in data collection section. Hence, in order to improve the SNR of the
particles, many identical looking particle images are aligned and summed (clustering) thus effectively increasing the SNR and dose without increasing the damage
[62]. There are three main advantages of reference-free (unsupervised) 2D classification: (i) to select few 2D classes from which we can make starting 3D map,
which can be projected as references for refinement. (ii) We can identify the fraction
of bad classes (which may contain artifacts, invalid particles, or simply empty), and
thus, those images with anomalies can be deleted from the data set in the beginning
itself. (iii) It also helps in identifying the conformational and compositional variability in the data set [50]. Two-dimensional (2D) and 3D classifications are carried
out by using various statistical analysis software suite IMAGIC [51], Spider [63],
EMAN [44], RELION-3 [64], FREALIGN [65], Appion [39], cryoSPARC [56],
ASPIRE Suite [58], Xmipp [30], SPHIRE (sphire.mpg.de), etc., or a combination of
more than one of these suites. Several of these software packages are integrated into
one processing framework, for example, as in Scipion [66]. An exhaustive list of
EM software programs is available at EMDataBank (EMDB, http://www.
emdatabank.org/emsoftware.html).
Spider [63, 67] and IMAGIC [51] were among the first programs to be developed for single-particle reconstruction in the year 1996 followed by FREALIGN
Single-Particle cryo-EM as a Pipeline for Obtaining Atomic …
385