Topaz training from 2D Class avg or 3D autorefine particles will work better?


I want an opinion regarding topaz training. Can we choose particles from the clean homo/hetero refine jobs from the sets of micrographs for the training purpose ? I have a decent reconstruction and these particles have already gone through extensive 2D cleaning previously.

In the tutorials I have seen people mostly selecting particles from the clean 2D class averages from different orientation. I just thought why not pick particles from the 3D job for the training if I already have a decent reconstruction.

Will there be any difference due to this ?

PS. Fun Fact. The name ‘Topaz extract’ is confusing as we already have a ‘extract from the micrograph’, rather topaz picking looks more appropriate. 1. ‘Topaz train’ for training the model then 2. Topaz picking for using that model to pick more particles from larger sets of micrograps 3. Then extract from the micrographs to extract these particles. Maybe the developers had something in their mind for these names :wink:

As I am a new user I found relion’s nomenclature more logical.


Hi Faisal,

It is fine to use particles from a 3D refinement for Topaz training. However, I would recommend still manually checking the particles you want to use using Manual Picker. Even in a nominally clean set of particles, there can be some remaining junk, and Topaz will pick what you train it to pick!


1 Like