Release notes

From Relion
Jump to navigation Jump to search

Below are the main new features for each release.

Release 3.0 and beyond

Please refer to releases pages in our GitHub repository.

Release 2.1 (June 2017)

New features compared to release 2.0

  1. Implementation of the Stochastic Gradient Descent algorithm for initial 3D model generation (like the one published by Marcus Brubaker).
  2. Functionality to impose helical symmetry in sub-tomogram averaging.
  3. Functionality to impose local symmetry, much like non-crystallographic symmetry in X-ray crystallography (publication in preparation).

Release 2.0 (January 2017)

New features compared to release 1.4

  1. GPUs can be used for the most demanding calculations (relion_refine and relion_autopick). One GPU card can equal the computational capacity of approximately 80-100 CPU cores, so this will probably transform the way you process images (or at least how much it costs). This work was done in the group of Erik Lindahl at the Scilife lab in Stockholm, Sweden. PhD students Bjorn Forsberg and Dari Kimanius did most of the hard work. A paper describing these benefits has been published in eLife, which also details the specific methods employed to use the GPU hardware architecture for highly parallel tasks.
  2. Functionality to process helical assemblies. This work was done by Shaoda He, a PhD student in the group of Sjors Scheres at LMB. The workflow of helical processing closely follows that of single-particles, so existing relion users should find this familiar. New parameters and tricks are explained on the Helical processing page. A paper about this work has been submitted to J.Struct. Biol., and is meanwhile available on bioRxiv.
  3. A new GUI which acts as a pipeline for image processing has been made. This will automatically log everything you do, and allow on-the-fly image processing during data acquisition. Details on how to use this are in the new tutorial, and a dedicated manuscript was accepted in the 2016 CCPEM Spring symposium special issue of Acta Cryst D.

Bug fixes during beta-testing

  • Using cross-correlation for all iterations resulted in segfault and termination.
  • Using cross-correlation during first iteration using GPUs returned the first orientation rather than the optimal orientation. The cross-correlation metric is used for the first iteration when the refinement option "Ref map is on absolute greyscale" is set to "no". This bug may thus affect early stages of classification/refinement, but is unlikely to cause distortions of final results.
  • Values of the ctf identical to zero resulted in program termination.
  • Input phase-shift was not accounted for.
  • When extracting particles into a directory already containing \_extract.star files which are not over-written, e.g. during re-extraction of a subset, the resultant particles.star contains all particles, possibly with conflicting information. This can occur if data was imported into the pipeline from outside the project directory. In this case, there will be a discrepancy between the number of particles selected and re-extracted. The conflicting meta-data may cause spurious errors during subsequent use of the re-extracted particles. Importing data from outside the project directory is still not intended and will cause re-extraction to over-write previous extracted data. The resultant particles.star will however now contain only the data selected for re-extraction.

New features including during beta-testing

  • Helical processing is now supported on GPUs.
  • Input data in 3D format (tomographic data) is now supported on GPUs.
  • Installation is now possible through apt-get. For instructions, see later sections.
  • Writing FOM-maps for autopicking is possible using several MPI-ranks. Wrinting such for more than 30 micrographs require manual intervantion to prevent accidental, heavy I/O loads.

Release 1.4 (September 2015)

  1. Full classification and auto-refine capabilities for sub-tomogram averaging (Bharat et al, Structure 2015)
  2. Focused classification with signal subtraction (see preprint on bioRxiv)
  3. 10x faster movie processing, and more efficient threads implementation
  4. Possibility to compile in single-precision (saves 50% RAM)

Release 1.3 (June 2014)

  1. a movie-processing procedure for small (sub-Megadalton) particles (Scheres, eLife 2014)
  2. a template-matching based (semi-)automated particle picking procedure (Scheres, J. Struct. Biol. 2015)
  3. a new GUI with image-displaying functionalities. Just click your way through the entire image processing pipeline.

Release 1.2 (June 2013)

  1. a statistical movie-processing procedure (Bai et al, eLife 2013)
  2. a semi-automated post-processing procedure (for improved resolution estimates after masking, MTF correction and B-factor sharpening; Chen et al, Ultramicroscopy 2013 )

Release 1.1 (September 2012)

  1. convenient preprocessing procedures
  2. a fully automated refinement procedure (3D auto-refine) for homogeneous data sets, employing "gold-standard" FSC calculations (Scheres&Chen, Nat. Methods 2012, Scheres, J. Struct. Biol. 2012)

Release 1.0 (November 2011)

  1. 2D and 3D classification procedures. The latter can also be used for high-resolution refinement (Scheres, J. Mol. Biol. 2012)