Job: Extract from Micrographs

Extract picked particles from micrographs using a specified box size.

Description

Extract picked particles from micrographs using a specified box size.

There are two job types available in CryoSPARC for Extraction. One runs on the GPU and the other on the CPU. Generally, extraction of particles is not computationally intensive but is bottlenecked by speed of reading and writing micrograph and particle stack files. It therefore runs equally fast on CPU compared to GPU in most cases.

The GPU-enabled job can utilize multiple GPUs, and the CPU-enabled job can utilize multiple CPU cores.

In CryoSPARC v4.0, the CPU extraction job has a new option to output two different particle outputs, one at a larger and one at a smaller downsampled box size. The smaller downsampled particles can be used for faster initial 2D and 3D classification, and the larger box size particles can be used for further processing and final refinements. See the Second (small) F-crop box size (pix) parameter description below.

Input

  • Aligned/motion-corrected micrographs

  • Particle locations

Common Parameters

  • Extraction box size: This controls the size of the extraction box, in pixels. See the Box sizes that allow for efficient processing section for appropriate sizes.

  • Fourier-crop to box size (pix) : If this is set, the job will also perform particle downsampling to this specified box size, using Fourier cropping. See the Box sizes that allow for efficient processing section for appropriate sizes.

  • Second (small) F-crop box size (pix) : This is only available in CryoSPARC v4.0+ and in the CPU extraction job type. If set, the job will output a second particles output called “Particles small” with the same particles but downsampled to a smaller box size using Fourier cropping. Connect these particles to downstream 2D classification or 3D reconstruction jobs, and later in the processing pipeline, you can use the “Particles” output’s blob result to replace the particles with the higher resolution extracted particles for refinements and further advanced processing. See the tutorial on low-level results. See the Box sizes that allow for efficient processing section for appropriate sizes.

  • For the GPU-enabled job type, you can set the Number of GPUs parameter to parallelize over GPUs.

  • For the CPU-enabled job type, you can set the Number of CPU cores parameter to parallelize over multiple CPUs.

  • Recenter using aligned shifts

    • Activated by default, this will cause the extracted picks to be re-centered if the 2D or 3D shifts are present from previous 2D classification or 3D reconstruction/refinement jobs

    • Re-centering will update both the location results (to store the new extraction centres) and alignments2D or alignments3D results (to store the residual <1 pixel shift after updating the locations)

  • Save results in 16-bit floating point: Saves the output particles in float16 format (CryoSPARC v4.4+).

Output

  • Particles

  • Particles small (Optional depending on parameters)

Notes and Limitations

  • Do not extract particles from micrographs if you plan to run Local Motion Correction, as this latter job type will perform extraction.

  • Particles too close to the edges of a micrograph will not be included in the particles output from this job. See the job stream log for more information on how many particles were not included.

Box sizes that allow for efficient processing

When selecting a box size or a Fourier-cropping box size, an even number must be selected. In addition, it is computationally more efficient to select numbers from the following list:

32, 36, 40, 42, 48, 56, 60, 64, 70, 72, 80, 84, 90, 96, 100, 108, 112, 120, 128, 144, 160, 180, 192, 200, 216, 224, 240, 256, 270, 288, 300, 320, 324, 336, 384, 400, 432, 448, 450, 512, 576, 640, 648, 672, 720, 768, 784, 810, 864, 882, 1024, 1152, 1280, 1296, 1344, 1440, 1568, 1620, 1728, 1792, 2000, 2048, 2160, 2592, 2744, 3456, 4096

The reason these box sizes are faster is that they have a small number of unique prime factors (2, 3, 5, 7, etc) and this is important for the Fast Fourier Transform to perform efficiently.

Last updated