2D classification

Home > 2D classification

2D reference-free classification

Unlike 3D analysis, which requires an initial model as a starting point, 2D classification of single particle EM data can be run without an 'initial model.'

2D classification results are a way to visualize how your 2D images will cluster together into homogenous class averages without any model required.

1. Methods for 2D classification
2. How does 2D classification work?
3. 2D classification to remove 'junk'

Methods for 2D classification

Generally, for negative stain data analysis, the neural network ('Auto Align') approach is fast and robust for calculating 2D classes from single particle data:

• Auto Align (Neural network)

Then, when performing CTF-correction for 2D classification, Relion's 2D classification routine provides high-quality 2D class averages:

• Relion 2D classification (Maximum likelihood)

How does 2D classification work?

2D reference-free analysis is important to verify the underlying structure of the proteins you are studying. In order to have confidence in the resulting averages, you need to have a robust alignment program that will give you realistic results each time.

In general, 2D reference-free analysis uses an iterative approach:

1. Classification (Multi-variate statistical analysis, Correspondence analysis, Topology representing network, Neural network) to create references
2. Align references together
3. Align particles to references
4. Repeat, with decreasing number of references

This means that over the course of 8 - 12 rounds, particles are grouped based upon similarities and then the data are re-aligned, and this process is repeated.

2D classification to remove 'junk'

After picking your particles automatically, you may need to 'clean up' your particle dataset in order to remove all of the 'bad' particles (e.g. stain/ice crystals, hole edges, etc.).

The best way to remove 'junk' is to run a first round of Auto_Align.py or Relion 2D classification, where you 'throw out' the particles that went to class averages that do not look like your sample.

To learn how to remove 'junk', read more here.