TOPAZ extraction failed due to particle threshold

Particle Picking
1

Hi everyone,

While using TOPAZ extract I sometimes get the error below.

  assert particles_extracted, "No particles extracted. Possible causes include \'Particle threshold\' parameter being too high or a version discrepancy between the Topaz or PyTorch instance that trained the model and the one that is being used for extraction."
AssertionError: No particles extracted. Possible causes include 'Particle threshold' parameter being too high or a version discrepancy between the Topaz or PyTorch instance that trained the model and the one that is being used for extraction.

This happens on some trained models but not others, so I think the TOPAZ version/pytorch are okay but rather something is not great with that particular topaz train job.

I am wondering: how many micrographs/particles should I be using to train TOPAZ? I used 250 images with 5000 particles total. These are the best particles after lots of classification. However for a different particle I used 800 microgaphs and 15000 particles and that did not give the TOPAZ extract error. However it only found very few particles. In both cases the training seems to have worked well as per the explanation of trainig on the cryosparc topaz page.

The particle threshold is already at 0, so I’m a bit lost as to what to do. Can I put the threshold to negative values?

2

5000 good particles should be plenty (1000-2000 is often fine).

You can reduce the threshold to negative values (I often use -2), but I would be surprised if you are not extracting any at a threshold of zero. Perhaps double check the extraction radius in case there is an issue there?

Cheers
Oli

3

thanks Oli

the radius of extracted region is default to 7 here. I guess this is downsampled 4X (original pixel size is 1.06 x 4 = 4.24A, X7 = ~29A). So my understanding is that it will not extract any particles within 29A radius of one another… so that seems reasonable. Or am I missing something?

thanks
Jesse

4

No that seems reasonable you’re right… hmm - not sure what the issue might be… @alexjamesnoble?

1 Like
5

Hi Jesse,

What value did you put for the ‘Expected number of particles’? If this value is more than ~50% away from the actual value, then the particle scores will be highly skewed. Try playing with that value if it is far off.

Another thing you can try: For Topaz Extract, set the ‘Particle threshold’ to -6.

Lastly, there are several ‘radius’ options in Topaz Extract. Make sure you only change the ‘Radius of extracted regions’ option. 7 is a reasonable number.

Let me know what the results are.

Best,
-Alex

1 Like