Topaz vs cryoSPARC coordinate wrangling query

Hi all,

After initially processing a dataset at bin1, I want to go back and reprocess a subset at bin2 (for the purposes of generating data for a workshop).

I redo Patch Motion with Fourier cropping to 1/2, redo Patch CTF, and re-extract after reassigning the appropriate particle set to the micrographs - so far so good, and the re-extracted particles look properly centered and give good 2D classes and a high resolution reconstruction.

However, when I attempt to use these particles to train a Topaz model, it complains that many (more than half) of the particles lie outside the bounds of the micrograph, and the resulting model is of course garbage.

So is there something different about the way that cryosparc is dealing with scaling particle coordinates internally, vs what is passed to Topaz? Or what is the reason?

Also, in the cryoSPARC wrapper for Topaz extract, it would be useful if the extraction radius could be specified in Å, rather than downsampled pixels, as in my experience this a point that is non-obvious and tends to cause some confusion.


Hi @olibclarke,

Thanks for reporting, and apologies for the (long) delay!. We suspect this is due to an issue with particle metadata not being updated properly. If you still have access to this data, would you be able to share the cs files of the particles and micrographs going into Topaz training? I can provide credentials for upload to our servers if you no longer have them saved from previous data transfers.


Hi Michael - I should do but it’s been a while, I will check! I should be able to reproduce though

1 Like