Job: Homogeneous Refinement (Legacy)

Homogeneous refinement (Legacy).


Rapidly refine a single homogeneous structure, using a particle stack an initial reference, to high-resolution and validate using the gold-standard FSC. Note: An updated version of this job is available, Job: Homogeneous Refinement (NEW), that supports higher-order CTF aberrations and optimized GPU kernels.


  • Particles

  • Initial model

  • Mask (optional)

Common Parameters

  • Refinement box size (Voxels): This determines the size of the internally stored volume. If this is null, the full particle size is used, otherwise images are automatically downsampled. This can be decreased to reduce memory requirements and increase speed, at the expense of omitting higher-resolution information.

  • Symmetry: This is the symmetry string corresponding to the enforced point-group symmetry to use during refinement. The allowed point group symmetry enforced are:

    • Cyclic (Cn, e.g. C2, C6, etc.)

    • Dihedral (Dn, e.g. D4, D7, etc.)

    • Tetrahedral (T)

    • Octahedral (O)

    • Icosahedral (I)

  • Re-estimate greyscale level of input reference: This can be enabled to estimate the multiplicative factor that correctly re-scales the reference to match the greyscale of the images. It is recommended to keep this on, especially if the input references are imported from elsewhere rather than generated within CryoSPARC.

  • Number of extra final passes: Normally, refinement terminates when the gold standard FSC resolution stops improving. However, extra iterations can be forced by setting this parameter to the desired number of extra iterations.

  • Maximum align resolution (A): This can be set to limit the resolution of information used for alignment. By default, information up to the gold standard FSC resolution is used for alignment, but higher resolution information can be ignored by setting this parameter. This can occasionally help prevent the build up of high-frequency overfit noise.

  • Initial lowpass resolution (A): The resolution to which the initial reference is lowpass filtered can be controlled by setting this parameter to the desired resolution in Angstroms. For smaller molecules, it may help to decrease this value to around 20 Å.

  • GSFSC split resolution (A): The resolution to which each half-set is considered independent can be controlled by setting this parameter.

  • Minimize over per-particle scale: This enables estimation of per-particle scale factors during Expectation-Maximization. This can help if ice thicknesses varied greatly during data collection. The iteration at which per-particle scale estimation begins can be controlled by setting the Scale min/use start iter parameter to the desired iteration number.

  • B-Factor for plotting: This sets the b-factor used during map sharpening and plotting. Negative values sharpen the structure, and positive values blur the structure. By default, Guinier estimation determines the b-factor used at each iteration.

  • Computational minibatch size: This controls the number of images in each minibatch, which influences the amount of memory that needs to be allocated for each batch. If GPU memory is limited, the minibatch size can be decreased to prevent memory allocation errors; if there is ample GPU memory, this could be increased.

  • Mask (dynamic, static, none): This controls the type of masking to use. This can be set to the following options:

    • dynamic, in which case a dynamic mask will be generated based on the reference at each iteration according to the dynamic masking parameters below.

    • static, in which case an initial mask must be provided, and it will be used unmodified at each iteration.

    • none, in which case no mask will be used during the refinement.

  • Dynamic masking parameters:

    • Dynamic mask near (A): This controls the distance (away from the thresholded reference) at which the mask is dilated. At the near distance, the mask value is 1.0.

    • Dynamic mask far (A): This controls the distance (away from the thresholded reference) at which the mask is softly reduced to 0.0.

    • Dynamic mask start resolution (A): When the map reaches this resolution (in Angstroms), dynamic masking will start.

  • Cache particle images on SSD: Particle images can be copied to the SSD for much faster read times during refinement, which dramatically speeds up the job. If this is enabled, particle images will remain cached on the SSD for future jobs using the same particle stack and running on the same node.


  • Refined 3D map

  • Half-maps

  • Mask used in refinement

  • Mask used in FSC calculation

  • Gold-standard FSC curve

  • Plots, including orientation distribution


For higher-order CTF aberration support, as well as faster codepaths and greater memory efficiency, refer to the Homogeneous Refinement (NEW) job, which subsumes the features of this job.

Common Next Steps

  • Download and inspect map

  • Launch a Sharpening Tools job to sharpen the volume with different b-factors

  • Launch a Create Templates job to generate templates from the reference, and iterate particle picking using the Template Picker

Last updated