New tools for automated high-resolution cryo-EM structure determination in RELION-3.

Zivanov J, Nakane T, Forsberg BO, Kimanius D, Hagen WJ, Lindahl E, Scheres SH. · 2018

Read New tools for automated high-resolution cryo-EM structure determination in RELION-3. by Zivanov J, Nakane T, Forsberg BO, Kimanius D, Hagen WJ, Lindahl E, Scheres SH. (2018) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc6250425.txt.

Introduction

Zivanov and co-authors describe the third major release of the RELION software. Their paper appeared in eLife. RELION builds three dimensional structures from electron microscope images. The method is called single particle cryo-EM. Version 3 adds speed from standard computer chips as well as graphics cards. That mix frees users from memory limits on one card type.

New tools pick particles from images with no starting template. Jobs can run from scripts while images are still being collected. Fine corrections handle particles sitting at varied depths in ice. Curvature of the imaging sphere is also corrected for large particles. Tests on public data sets gained clear gains in detail over older versions. The gains ranged from 0.2 to 0.7 ångströms. Such steps help labs reach atomic level models faster.

Catalog record

CREATORZivanov J, Nakane T, Forsberg BO, Kimanius D, Hagen WJ, Lindahl E, Scheres SH.
YEAR2018
COLLECTIONPAPERS
FORMATTXT
FILE SIZE75 KB
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC6250425

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

New tools for automated high-resolution cryo-EM structure determination in RELION-3 By Jasenko Zivanov, Takanori Nakane, Björn O Forsberg, Dari Kimanius, Wim JH Hagen, Erik Lindahl, Sjors HW Scheres, Edward H Egelman, John Kuriyan Published in: eLife DOI: 10.7554/eLife.42166 ABSTRACT Abstract Here, we describe the third major release of RELION. CPU-based vector acceleration has been added in addition to GPU support, which provides flexibility in use of resources and avoids memory limitations. Reference-free autopicking with Laplacian-of-Gaussian filtering and execution of jobs from python allows non-interactive processing during acquisition, including 2D-classification, de novo model generation and 3D-classification. Per-particle refinement of CTF parameters and correction of estimated beam tilt provides higher resolution reconstructions when particles are at different heights in the ice, and/or coma-free alignment has not been optimal. Ewald sphere curvature correction improves resolution for large particles. We illustrate these developments with publicly available data sets: together with a Bayesian approach to beam-induced motion correction it leads to resolution improvements of 0.2–0.7 Å compared to previous RELION versions. Research organism: None INTRODUCTION Macromolecular structure determination by single-particle analysis of electron cryo-microscopy (cryo-EM) images has undergone rapid progress in recent years. Before 2010, cryo-EM structures to resolutions beyond 4 Å had only been obtained for large icosahedral viruses, for example Jiang et al., 2008, Yu et al., 2008, and Zhang et al., 2008. Since then, cryo-EM structures to resolutions that are sufficient for de novo atomic modelling have been achieved for a wide range of samples (Fernandez-Leiro and Scheres, 2016), and several structures beyond 2 Å resolution have been reported (Merk et al., 2016; Bartesaghi et al., 2018). In 2012, the availability of the first prototypes of direct electron detectors, which recorded images with unprecedented signal-to-noise ratios (McMullan et al., 2009), represented a crucial step forward. This advance was closely followed and partially overlapped with important improvements in image processing software. Together, they led to what was termed the ’resolution revolution’ in cryo-EM structure determination (Kuhlbrandt, 2014). Since the start of this revolution, many new software solutions have been introduced. Some new programs exploited specific opportunities provided by the new detectors, like the correction of beam-induced motions through movie processing (Brilot et al., 2012; Li et al., 2013; Rubinstein and Brubaker, 2015; Abrishami et al., 2015; McLeod et al., 2017; Zheng et al., 2017). Other programs tackled individual steps in the single-particle workflow, like fast and robust estimation of the contrast transfer function (CTF) (Rohou and Grigorieff, 2015; Zhang, 2016), local-resolution estimation (Kucukelbir et al., 2014), or on-the-fly processing of images while they are being acquired at the microscope (de la Rosa-Trevín et al., 2016; Biyani et al., 2017). In the mean time, also general-purpose programs that already existed, like EMAN2 (Tang et al., 2007), Frealign (Grigorieff, 2007) or IMAGIC (van Heel et al., 1996), were further improved and used for high-resolution structure determination. Both old and new programs benefitted from the increase in image signal-to-noise ratios provided by the new detectors. The availability of the first direct-electron detectors coincided with the introduction of the RELION software (Scheres, 2012b). This computer program introduced an empirical Bayesian approach, in which optimal Fourier filters for alignment and reconstruction are derived from the data in a fully automated manner (Scheres, 2012a). This strongly reduced the need for user expertise in tuning ad-hoc filter parameters compared to existing programs, and probably contributed to the rapid uptake of cryo-EM structure determination by non-expert groups. Nowadays, RELION is widely used (Patwardhan, 2017), and similar statistical approaches have been implemented in software packages that were developed since: cryoSPARC (Punjani et al., 2017) uses the same regularised likelihood optimisation target as RELION, and SPHIRE (Moriya et al., 2017) and cisTEM (Grant et al., 2018) implement likelihood-based classification methods. RELION underwent a major change in 2016 with the release of its second version, which comprised a pipelined approach for the entire single-particle workflow (Fernandez-Leiro and Scheres, 2016) and used acceleration on graphical procesing units (GPUs) to reduce its computational costs (Kimanius et al., 2016). In addition, its functionality was expanded with the incorporation of sub-tomogram averaging (Bharat et al., 2015) and helical reconstruction (He and Scheres, 2017). In this paper, we introduce version 3 of RELION. In the Materials and methods section, we describe algorithmic improvements and changes in implementation that have been made throughout the single-particle pipeline in RELION-3. In the Results and Discussion section, we illustrate how these changes allow for improved automation, fast accelerated reconstruction on normal hardware (CPUs) in addition to GPUs, and how the new features enable higher resolution reconstructions compared to previous releases. MATERIALS AND METHODS Laplacian-of-Gaussian auto-picking The recommended procedure for the selection of particles from micrographs in previous versions of RELION was to calculate templates for reference-based auto-picking from particles that were manually picked in a subset of the micrographs (Scheres, 2015). To allow non-interactive, on-the-fly processing, we implemented a reference-free particle picking method based on the Laplacian-of-Gaussian (LoG) filter. Our method is related to the approach in DoGpicker (Voss et al., 2009), which uses a Difference-of-Gaussian (DoG) filter. Similar to the implementation in swarmPS (Woolford et al., 2007), we apply a LoG-filter with an optimum response for blobs of size d by multiplying the Fourier transform of a micrograph by [formula: (1)|k|2σ2⁢exp⁡(|k|2-2⁢σ2)] where k∈ℝ2 is the image frequency, σ=2/d, and the division by σ2 outside the exponentiation provides a normalised response of the filter for different values of d. In order to deal with elongated particles, the user specifies a minimum (dm⁢i⁢n) and maximum (dm⁢a⁢x) particle diameter. The program then applies four LoG-filters to detect areas in the micrograph with blobs that are smaller than dm⁢i⁢n (with d=dm⁢i⁢n/n and n=2,3,4 or 5); four filters to detect blobs that are larger than dm⁢a⁢x (with d=n*dm⁢a⁢x and n=2,3,4 or 5); and three filters to detect blobs within the user-specified particle size (with d=dm⁢i⁢n, d=(dm⁢i⁢n+dm⁢a⁢x)/2 and d=dm⁢a⁢x). Only pixels for which the highest LoG-filtered value over all eleven filtered micrographs (L⁢o⁢Gb⁢e⁢s⁢t) is above a user-controlled threshold and the corresponding blob size db⁢e⁢s⁢t is in the user-specified particle diameter range are considered for particle picking. Particles are picked iteratively at the highest remaining L⁢o⁢Gb⁢e⁢s⁢t value, and around each picked particle one discards all pixels within a circle with diameter db⁢e⁢s⁢t. This process is repeated until no suitable pixels remain. To normalise the threshold for L⁢o⁢Gb⁢e⁢s⁢t over many different data sets, the default threshold is set to the mean of all L⁢o⁢Gb⁢e⁢s⁢t values for which the corresponding db⁢e⁢s⁢t is in the user-specified particle diameter range, and deviations from the default are expressed in standard deviations of those same values. CTF refinement In RELION, per-micrograph CTF parameters are determined through wrappers to CTFFIND (Rohou and Grigorieff, 2015) or Gctf (Zhang, 2016). These programs fit CTFs to the Thon rings visible in the power spectra of (patches of) micrographs. In RELION-3, we have implemented a program to refine the CTF parameters, that is to re-estimate defocus and astigmatism, using a 3D reference structure. This allows CTF estimation to exploit both the phases and the amplitudes of the experimental images, instead of having to rely exclusively on their power spectra. This approach thus produces significantly more reliable estimates, due to higher signal-to-noise ratios and also because it does not require separation

Read the complete text