Skip to content
No description, website, or topics provided.
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
.gitignore
CMakeLists.txt
LICENSE
README.md
compile.sh
fouriers.c
fouriers.h
interact.c
lafter.c
lafter.h
realspace.c
realspace.h
suppress.c
suppress.h
truncate.c
truncate.h
upsample.c
upsample.h

README.md

LAFTER [ Local Agreement Filter for Transmission EM Reconstructions ]

KAILASH RAMLAUL, COLIN PALMER & CHRISTOPHER H S AYLETT - 14/11/2018

TL-DR:

  • LAFTER is a local filter for single particle TEM reconstructions
  • LAFTER requires FFTW3 compiled with POSIX threads for compilation
  • LAFTER is a C99 program (a compilation script is included)
  • LAFTER prints all its available options to stdout when called
  • LAFTER requires independent half-volumes and mask from refinement
  • If no explicit mask was used give diameter to which noise remains
  • LAFTER reports the Cref and LAFTER-sum xFSC curves as validation
  • LAFTER outputs a twofold upsampled MRC volume for interpretation
  • The output of LAFTER is INCOMPATIBLE with coordinate refinement
  • Overfitting will occur if this is tried as the maps are denoised

LAFTER Algorithm

  • LAFTER requires the independent unfiltered halfmaps for agreement estimation, and any and all masks used for the refinement process to identify regions where the noise has been down-weighted.
  • LAFTER performs two passes over the volume, the first from low to high resolution for noise minimisation, and the second from high to low, to filter each voxel at the point of agreement.
  • The first pass over increasing resolution suppresses noise in each halfmap by weighting each voxel in each shell by the probability that it is part of the noise distribution. This can be thought of as “adaptive masking”.
  • During a second pass the two noise-weighted half-maps are combined at decreasing resolution and each voxel lowpass filtered according to the maximum value of the observed noise distribution to attempt to avoid the carry-over of any noise into the final volume.
  • Each successive resolution shell is incorporated using an adaptive step size proportional to both the current resolution, and the SNR between halves, minimising the step size at low resolution, where the signal varies strongly, and near the resolution-limit, as the noise eclipses the signal.
  • The incorporation of new resolution shells terminates when the FSC between the masked halfmaps reaches either 0.143 (default), or the user provided FSC cut-off ( --fsc fsc_cut-off e.g. 0.5 ).
  • The final density maps are then explicitly lowpass filtered at the resolution cut-off to truncate the resolution at this point.

Calculation of CRef and comparison to LAFTER-density FSC

  • The theoretical value of CRef is calculated between the unfiltered halfsets according to the equation of Henderson and Rosenthal: double Cref = sqrt((2 * FSC) / (1 + FSC))
  • Here the FSC refers to the half-set FSC. Cref is compared to the cross FSC (xFSC) between the summed dataset and the output of the LAFTER algorithm.
  • All FSCs are calculated with masked densities in every instance.

Installation and License

  • LAFTER is a performance optimised C program using FFTW3 in Fourier transformation (Frigo and Johnson, 2005) for speed and portability
  • Compilation is easiest with CMake. Build and run LAFTER with these commands:
mkdir build && cd build
cmake ..
make
./lafter
  • LAFTER is open source and is made available under the GNU public license, which should be included in any package.

    ++++ Good luck & happy cryo-EM-ing - Kailash, Colin & Chris ++++

You can’t perform that action at this time.