EMAN2 Release Notes

• Major changes and improvements to e2gmm
• New EMAN3 pipeline being prototyped, including e3make3d_gauss and incorporation of Gaussian reconstruction into e2spa and e2spt refinement programs.

With Pawel Penczek's retirement, SPHIRE and SPARX are no longer actively developed or maintained

• Full PPPT subtomogram averaging pipeline capable of near-atomic resolution
  • Numerous improvements for in-situ subtomogram averaging
  • New dedicated deep-learning tomography particle picker
  • Simultaneous (visual) picking of multiple macromolecular species
  • New options for tilt series alignment in tricky cases
• Data compression support for both CryoEM and CryoET
  • Native HDF5 mechanism
  • lossless or lossy with user-selectable bit count
  • Files still readable by Chimera
  • No quality/resolution loss
  • Typical 10-20x size reduction, dramatically improves disk I/O and network transfer
• Extensive work on e2boxer
  • improved deep learning picker
  • Fixed problems with reference-based and local pickers
  • Added simple inline instructions
• Deep learning GMM for single particle variability studies (e2gmm.py, see arxiv paper, experimental)
• New local resolution and filtration based on mFSC (Penczek), for SPA and SPT (iterative)
• New visualization options for volume stacks
• Support for EER format & oversampling (counting-mode Falcon 4 images)
• Colored isosurfaces functioning, and automatic local resolution display when computed
• Speed improvements for image I/O operations
• Better integration with Numpy and Jupyter image object data sharing (advanced users)
• Improved option for x/y/z projections in e2display
• Drag and drop support for rearranging images in tiled image display
• Python 3 based (finally!)

• General changes:
  • New browser options for display of stacks of 3-D volumes
  • RCTboxer works properly again
  • Fixed a problem reading A/pix from FEI-style MRC files
  • New Processor for bit-compression of image files (makes images more compressible)
• PPPT Subtomogram averaging:
  • Automatic CTF based handedness checking for tomograms
  • Focused refinement https://blake.bcm.edu/emanwiki/EMAN2/e2tomo_more#Focused_refinement
  • Better parallelism
  • Slightly improved tilt series alignment (resolution improvement)

• Support for processing helical specimens
• AutoSPHIRE - automatic refinement tool
• Integration of crYOLO
• Cinderella for 2D class selection

• A complete CryoET pipeline from tilt series through subnanometer resolution hybrid subtomogram averaging
  • fiducial-less fully automated tilt series alignment (also works with fiducials)
  • rapid tiled Fourier reconstruction
  • full tilt/geometry aware CTF correction
  • multi class 3-D particle picker with new ties to deep-learning annotation
  • SGD automatic initial model generation
  • traditional 3-D subtomogram averaging
  • per-particle per-tilt hybrid subtomogram/single particle reconstruction to subnanometer resolution
• A new switchable filter in e2boxer making it dramatically easier to distinguish particles in images
• New improvements to bispectral classification for 2-D unsupervised classification and 3-D refinement
• Focused classification in 3-D refinement available from the workflow interface
• Improvements to e2extractsubparticles as an alternative to focused classification
• Upgrade from Qt4 to Qt5 as part of the process of transitioning to Python3 over the next year
• Many minor bugfixes

• Fix bug causing “missing SNR” problem during refinement in specific situations
• Testing support for bispectrum based class-averaging and 3-D refinement. ~10-50x faster. Better 2-D class averages
• Better support for GPU in Neural Network tomogram segmentation and particle picking
• Direct support for phase plates in CTF correction with adjustable phase slider and autofitting (first version, room for improvement). Issues with astigmatism in this version.
• Better GUI display of CTF and Astigmatism
• e2symsearch3d bugs fixed
• Many subtomogram averaging bugs fixed. New pipeline under development.
• Many improvements to e2evalrefine for particle and class-average assessment

• Many, deep improvements to refinement
  • Substantial refinement changes and new filtering techniques
  • Optional tophat filter (similar to Relion post processing), side chains often (but not always) look even better than Relion/CryoSparc
• Local resolution and filtration
  • can be enabled in refinement to provide local detail appropriate to local resolution
• Several new methods for conformational/compositional heterogeneity
  • Including multi-model refinement with or without alignment, masked particle subtraction, 2-D and 3-D Deep Learning approaches (experimental)
• New bad particle identification strategy
  • Proven to produce better maps in several projects!
• Automatic CTF
  • Used to be several manual steps. Entire process now automated.
  • Easy and fast refinement at progressive resolutions within a single project
• Phase plate CTF correction
  • Supports phase shifts covering full 360 degree range, with explicit 'phase' slider
  • No automatic fitting of phase shift in this version (next minor release)
• New e2boxer (particle picker)
  • Fixes the problems with the old particle picker
  • New (optional) neural network picker for difficult projects
• Stochastic Gradient Descent initial model generator (experimental)
• Automatic magnification anisotropy correction tool
  • Post-processing program which corrects for the common microscope anisotropy problem on FEI scopes
  • Automatic, does not require additional data collection
• New direct detector movie aligner
  • All new program. Competitive with other alignment programs in quality
  • Workflow for handling movies in EMAN2 projects
• New localweight averager (experimental)
  • excludes “bad” parts of individual particles (overlaps, contamination, etc.)
• New 2-D registration algorithm
  • scales well with box size
  • more accurate going into “refine” alignment (in many cases refine can be skipped)

• New subtomogram averaging tools
  • New pipelines for subtomogram averaging and classification
• Up to 20x faster 3-D alignments,
  • now practical to study 10,000 300x300x300 particles on a single workstation
• New automatic missing wedge identification/compensation in alignment/averaging

• Workflow for semi-automatic tomogram annotation/segmentation
  • Uses convolutional neural network technology with user guided training of features.

• Anaconda Python based distribution
  • Integrates !SciPy, Theano, !PyLearn and other toolkits
• New installers, with better OpenMPI/Pydusa support
• !GitHub
  • Source code is now managed via a public !GitHub repository (cryoem/eman2)
• Windows 10/7 64 bit
  • Initial support, poorly tested, but available for the first time (EMAN2 only, SPARX/SPHIRE do not support this platform)