Interactive Job: Select 2D Classes

Description

Interactively select 2D classes from the output of 2D Classification job. If particles (from the same 2D Classification job) are included in the inputs, then particles will also be partitioned based on their assignments to the selected class averages.
The Select 2D Classes job most commonly follows any 2D Classification job in which there are one or more "junk" classes. Removing "junk" particles from the particle dataset is an important step in particle curation, and will help increase the quality of 3D reconstructions and refinements. Note that Select 2D Classes can also be used on the output for any job that creates templates (e.g. on the templates generated in a Create Templates job)

Input

  • Class averages
  • Particles (optional: must be outputted from the same 2D Classification job as the class averages)

Common Parameters

  • Auto Thresholds: If you only want to select class averages that are better than a known resolution, and/or have particle counts greater than a known value, either of these thresholds can be set. Note that setting either threshold will skip the interactive process, so these thresholds should only be set if you are not interested in visually comparing and selecting the classes.
    • Classes where resolution better than: A resolution value in Angstroms; all selected class averages will have higher resolutions than this
    • Classes where particle count higher than: All selected class averages will have at least this many particles assigned to it

Output

  • Selected class averages
  • Selected particles
  • Excluded class averages
  • Excluded particles

How to use Select 2D Classes

Below is an example of the Interactive tab in a Select 2D Classes job, with 11 classes selected. At the top right of the panel, the number of currently selected particles, as well as the total number of particles in the dataset, are shown.
Each desired class can be selected by clicking on it
Click on each desired class to select it. To improve the quality of your particle dataset, avoid selecting classes that contain only a partial particle, two or more particles, or a non-particle junk image (e.g. ice crystals). You can use both the number of particles and the provided class resolution score to identify good classes of particles. There are several ways to sort the classes in ascending or descending order, shown along the top of the panel:
  • # of particles: Sort by the total number of particles in each class
  • Resolution: Sort by the relative resolution of all particles in the class (Å)
  • ECA: Sort by the number of Effective Classes Assigned (ECA)
In addition, for dealing with large numbers of classes, you may want to select all classes that meet a given threshold in particle count, resolution or ECA value. To do this, first choose a class that has the desired threshold value (in the image below, the desired threshold was particle counts greater than 29309 particles). Next, right-clicking on the chosen class will reveal a drop down menu with options to select all classes that meet or fail this threshold, as shown below.
Right-clicking on a class (in the second row) reveals the drop down menu for selecting classes that meet the given thresholds
The "Select All", "Select None", and "Invert Selection" buttons along the top of the panel can be used to quickly select all classes, clear all selections, and invert the current selection, respectively. These may be useful for dealing with a large set of class averages.

Common Next Steps

The most common next step is Ab-initio Reconstruction, to generate one or more initial models from the selected (curated) particles.
If you would like to increase the number of particles in your particle dataset by re-picking, you can use the selected class averages as templates for a Template Picker job.
If you would like to further clean your particle dataset, subsequent rounds of 2D Classification using the selected particles with varied number of classes to continue removing junk particles. As well, in cases where strong preferred orientation is biasing results of a 3D reconstruction or refinement, Rebalance 2D Classes (BETA) may be useful in order to cluster the class averages by similarity, and even out the number of particles in each cluster.
Finally, if an initial model is already available, any of the 3D Refinement jobs are also common next steps.
Last modified 1mo ago