12. 3D classification of particles

The duplicate removal job that we ran in the last step cannot remove all the bad particles. During picking, we approximated HIV capsids as spheres but they are really not. Also, some capsids are not complete. By approximating them to spheres and picking by sphere sampling, we include both true particles as well as some junk particles. Before we try high resolution refinement at bin 1 level, we need to remove those junk particles to achieve a good resolution. We do this cleaning up by running a 3D classification job on our previously refined and duplicate-removed particles.

To do this in Doppio, go to NEW JOB tab in Doppio GUI and type relion.class3d.tomo in the search bar. Then select the class3d job and enter inpust parameters. Our input parameters are given in the image below:

Note that in our classification we’ve lowpass filtered the reference to 60 A to avoid (minimise) the reference bias, as well as the symmetry was chose to C1 because junk particles may not necessarily have C6 symmetry. We encourage you to read the RELION documentation for details.

Relion 3D classification documentation.

12.1. Results

3D classification results can be found in Class3D/job015/ directory.

We have asked for 30 iterations, the class distribution scores (probabilities) at the end of the 30 cycles are plotted below. You will find them in the RESULTS tab.

Note

If you want to look at intermediate results as the classification progresses in Doppio, you can do this by clicking on the (…) of the job button and selecting the Recalculate results option.

Class001 and 002 are the two main classes came out at the end of classification. Out of these two class002 is the major with +0.55 probability. This class contains the most of the good particles, and the orthogonal slices of their average particle is shown in the top line of the right panel. The resolution of this class at iteration30 is 6.7 A as shown in the Class3D/job015/run_it030_model.star file. Below is a screenshot of the model file.

In this exercise, our main purpose of 3D classification was to obtain a particle set that are as homogenouse as possible, that will ultimately yeild us the best possible resolution. In your studies, the goals may be different. In the end, we dicided to move forward with class2 particles only, and there were 4630 particles.

12.2. Selecting the class002 particles

We used the ccp-em pipeliner class selection routine implemented in Doppio for this. To call this functionality, got to NEW JOB tab in Doppio GUI and type pipeliner.select.classes in the search bar.

Classification optimiser file is Class3D/job015/run_it030_optimiser.star and under the Classes to select type 2 and run the job. This will write out a new particle file. In our processing workflow it was Select/job017/particles.star.