Fundamentals of image reconstruction from projection images

8/13/18, Ludtke, VARI
Steve Ludtke
Charles C. Bell Professor
Biochemistry and Molecular Biology
Director, CryoEM/CryoET Core
Co-director CIBR Center
Baylor College of Medicine
Introduction to Single Particle Reconstruction
(and a quick intro to EMAN2 for the afternoon)

Biological TEM Methods
• Single Particle Analysis (SPA or SPR)

• Cellular Tomography (CryoET)

• Subtomogram Averaging

• Helical Processing

• 2-D Crystallography

• 3-D Microcrystals (MicroED)

190/~30,000 Particle Images

Bai, X. C., Fernandez, I. S., McMullan, G. & Scheres, S. H. Ribosome structures to near-atomic resolution from thirty thousand cryo-EM
particles. Elife 2, e00461 (2013). PMC3576727.

TEM Produces Projections

Single Particle Analysis
Extract Particles
Determine
Orientations
3-D
Reconstruction
Interpret Map
Movie Alignment
Record Images
Prepare Grids

Extract Particles
Determine
Orientations
3-D
Reconstruction
Interpret Map
Single Particle Analysis

7.5° Angular Step

Determine Particle Orientations
?
...
Every particle vs every possible projection?

✴ 30,000 particles

✴ 3000 projections (~2.6 degree sampling)

✴ 180 rotations

✴ 10 x 10 translations (if centering is decent)

✴ 65,536 pixels (256x256 image)

✴ 20 FLOPS/pixel

= 2.1 x 1018 FLOPS (2 exaFLOPS)

• @ 100 gigaFLOPS/s = 6,000 hours (~ 240 days on one typical workstation)

• Clearly we need to be smarter...

• (Actual EMAN2 time <3 hours on ... even more cleverness possible though)
(Ribosome at ~4 Å)

✴ 30,000 particles

✴ 3000 projections (~2.6 degree sampling)

✴ 180 rotations

✴ 10 x 10 translations (if centering is decent)

✴ 65,536 pixels (256x256 image)

✴ 20 FLOPS/pixel

= 2.1 x 1018 FLOPS (2 exaFLOPS)

• @ 100 gigaFLOPS/s = 6,000 hours (~ 240 days on one typical workstation)

• Clearly we need to be smarter...

• (Actual EMAN2 time <3 hours ... even more cleverness possible though)

Reconstruction Algorithms
• Back Projection

• Filtered Back Projection

• Direct Fourier Inversion

• SIRT

• SART

• ...

Reconstruction & the Asymmetric Triangle

FFT
Crystals have spots
Particles are ~continuous and have phases!

FFT

IFT

One problem...
• To determine the particle orientations, we needed a 3-D reference

• catch 22

• All current CryoEM Single Particle solutions are iterative!

Spider
(typical method, late 70s, Pub 1981)
Extract Particles
Determine
Orientations
3-D
Reconstruction
Interpret Map
Initial Model
Particles
Projections
Determine
Orientations
3-D
Reconstruction
Final
Reconstruction

Extract Particles
Determine
Orientations
3-D
Reconstruction
Interpret Map
Initial Model
Particles
Projections
Alignment
3-D
Reconstruction
Final
Reconstruction
Class
Averages
PCA
Common
Lines
IMAGIC
(typical method, 1981)

EMAN
(1998)
Extract Particles
Determine
Orientations
3-D
Reconstruction
Interpret Map
Initial Model
Particles
Projections
Classify
Particles
3-D
Reconstruction
Final
Reconstruction
Class
Averages

Extract Particles
Determine
Orientations
3-D
Reconstruction
Interpret Map
Initial Model
Particles
3-D
Reference
Determine
Probabilities
Max. Lik.
Reconstruction
Final
Reconstruction
Relion
(2012, adapted from XMIPP)

Model Bias
25 100 250 1000 2000
Align toNoisyBase

Model Bias
25 100 250 1000 2000
Align toNoisy (~10% contrast)Base

Model Bias
25 100 250 1000 2000
Align toBase Noisy (~10% contrast)

Model Bias
25 100 250 1000 2000
Align toNoisyBase
Iter x4

Model Bias
25 100 250 1000 2000
Align toNoisyBase
Iter x8

9/8/18, Ludtke, IMC19
4096 Particles of Noise
reﬁne 6 mask=56 hard=90 sym=d7 ang=1.6071 pad=160

xﬁles=2,800,99 amask=15,.9,16 phasecls classkeep=10 sep=3
How About 3-D ?

Initial Model 1 Iter. 2 Iter.
3 Iter. 4 Iter. 5 Iter.
no iteration

3 Iter. 4 Iter.
1 iteration

3 Iter. 4 Iter.
6 iterations

3 Iter. 4 Iter.
6 iterations

(8 A lowpass)

"Gold Standard" Refinement
Particles
Odd
Even
Initial Map
1
Refine
Refine
Initial Map
2
FSC Resolution
Final
Map
Filter
Reduces impact of noise bias on
resolution, but does not cure it!

EMAN2.1
320 CPU-hr
Relion 1.3
2200 CPU-hr
S. Cereviseae 80S Ribosome (EMD-2275)

Data taken from PDBe 3DEM test data
no movie alignment performed

Dataset 10002 (Bai XC, Fernandez IS,
McMullen G, Scheres SH)
Ludtke, 1/28/16

EMAN2 Relion
TRPV1
Liao, M., Cao, E., Julius, D., and Cheng, Y. (2013). Structure of the TRPV1 ion channel determined by electron
cryo-microscopy. Nature. 504:107-112.
3.7 Å

30 kDa HIV-1 RNA Dimerization Signal
~9 Å

30 kDa HIV-1 RNA Dimerization Signal
~9 Å~15 Å
Relion

EMAN2.2
http://eman2.org
EMAN2 and SPHIRE/SPARX share a common core and
are distributed together if you have one installed, you also have the other!
However, they are completely independent beyond that.
(this should also soon include Niels Volkmann's PyCoan)

History
• 1999 - EMAN1 released

• First to do Fourier-based reconstruction, full CTF correction, GUI

• 2003 - Started development of EMAN2

• Python Focus, redesign from scratch

• SPARX = EMAN2/PHENIX crossover, Pawel Penczek

• 2008 - Grant supporting SPARX ends -> Pawel changes its focus

• 2005 - 2009 Various pre-releases of EMAN2/SPARX

• 2010 - EMAN2.0/SPARX

• 2014 - EMAN2.1

• 2016 - SPHIRE - first beta

• 2017 - EMAN2.2
Relion first release
Frealign first release
Simple first release
CryoSPARC first release
CISTEM first release
~1980 SPIDER & IMAGIC

What can it do?
• Movie alignment (continued work in progress)

• Micrograph screening

• CTF (micrograph, particle, tilt series)

• Single Particle Analysis

• 2-D Variability

• 3-D Variability

• Tomography alignment and reconstruction

• Subtomogram averaging

• Tomogram Segmentation

• Utility Functions (ﬁle conversion, image processing,…)

EMAN2 Features
• Complete graphical workflow, all steps of single particle refinement
• Project system which organizes data and records all reconstruction info.
• Qt/OpenGL for 2d & 3d display.
• Support for all documented cryoEM file formats
• Over 200 general purpose image processing algorithms
• Interoperability features with Frealign, Relion, ResMap, IMOD…
• Tilt Validation, Random Conical Tilt, Single Particle Tomography
• Parallel processing using MPI and/or Threads

EMAN2 Architecture
C++ Core
Python Core
Command-Line Programs
High-Level Programs
Project Manager
Interface
Easy
Flexible

Extensible Core
Type Description #
Processor Generic image processing algorithms,
ﬁlters, masks, thresholds, etc. 220
Aligner
Algorithms used to align 2 images or
volumes to each other 32
Projector
Routines to generate 2-D projections of
3-D objects 7
Reconstructor
Routines to reconstruct 3-D objects from
2-D projections 13
Cmp
Similarity metrics used to compare two
images or volumes 15
Averager
Average together stacks of images in
various ways 12
Analyzer
Perform various operations on sets of
images, such as classiﬁcation or PCA 9
Orientgen
Routines describing how projections cover
the asymmetric triangle 7

MRC R/W IMAGIC R/W
SPIDER R/W HDF5 R/W
PIF R/W ICOS R/W
VTK R/W PGM R/W
Amira R/W Xplor W
Gatan DM2 R Gatan DM3 R
Gatan DM4 R FEI SER R
TIFF R/W Scans-a-lot R
LST R/W PNG R/W
Video-4-Linux R JPEG W
File Formats

Programs
• Command-Line Programs (EMAN2)

syntax:

e2<name>.py --help

e2<name>.py <ﬁle> [--option=value] [--option] [-O]

<> - required parameter

[] - optional parameter

• e2help.py <category>

Fundamentals of image reconstruction from projection images

Recommended

Recommended

More Related Content

Similar to Fundamentals of image reconstruction from projection images

Similar to Fundamentals of image reconstruction from projection images (20)

More from Dominika Elmlund

More from Dominika Elmlund (11)

Recently uploaded

Recently uploaded (20)

Fundamentals of image reconstruction from projection images