CytoKit for other multiplexing methods

[NieHauLab]

Dear cytokit-team,

recently I read your paper and found it really cool toolkit. We are also working with a multiplexing technique which is called "Multi Epitope Ligand Cartography" (MELC). I only would like to ask you, if cytokit could also be configurated in this way, that those images are processed, or at least usable in the cytokit explorer, because image alignment, background subtraction and shading correction is allready done...so at the end we just have to feed these final stack into a readable format for the cytokit explorer, right?

• Our raw images are registrated by taking a phase contrast image instead of nuclei staining
• image illumination correction is performed by using bleaching images of the previous and current staining

more information can be found here: https://onlinelibrary.wiley.com/doi/full/10.1002/cyto.a.23526

At the end we have a stack of every single staining...like multi channel image....most of the time, we only have one field of view, because the method is kind of limited to 4 field of views...so no tile scanning like in CODEX.

With the cytokit explorer I would like to give my colleagues another insight to their data. Handling and interaction in more than 80 stainings in one tissue sample is quite complex, as you maybe know.

Thank you very much in advance

[eric-czech]

Hi @NieHauLab,

Do you already have label images that define where the objects in the volumes are? You could use Cytokit Explorer just for visualization, but I think there are better tools for that unless you are trying to cross-filter with objects (and there may still be better options like Cell Profiler Analyst if that's all you need to do).

I may be able to help more if you could add detail about what insights you're trying to find. If I'm understanding you correctly though, then I don't think Cytokit will be of much use to you unless you have already done object segmentation or don't care about it -- and better options exist for the latter case.

[NieHauLab]

Hey Eric,

thanks for the answer...maybe you are right, trying at first CellProfiler Analyst for such data overview, so no extra configuration is needed. My colleagues are just a little bit frustrated by looking at the time CP needs for calculation and ask me, if there is something faster combined with nice GUI and direct data availability...because in principle the first thing is to get an idea of coexpression...then we go to neighborhood analysis and phenotyping tasks...but there we are at the beginning and looking for right tools...

By the way, do you have kind of giude to use CPA in this manner? I read, that data has to be saved in Database and is only then usable within CPA

[eric-czech]

Hm well if you plan on needing to program many of those tasks yourself, then Cytokit is a good start if you want to stay within the Python ecosystem. It doesn't sound like the specific GPU implementations in this package are necessarily relevant, but if you think CellProfiler is slow then I can't say I know of many faster options.

You do have to use CP tools to build a CPA database. There are at least ways to use those tools programmatically and though that is not supported officially by CP (it's on the roadmap), this project would give you a sense of how to do it. It supports mysql and sqlite databases that have a straightforward schema so there's no reason you can't use the data in them outside of CPA.

[NieHauLab]

hello eric,

I had a look to the CPA and if I saw right, then there are some things missing compared to your Cytokit explorer...even the selection tool is still affected by a bug and this is why the function is deactivated. Additional to this even if the selection tool would work, I don't know, if you could observe the selection inside the images.

Do you think there will be a stand alone version of the cytokit explorer in future, supporting own multiplexed data in a way, that user would only need segmentation mask, image data and feature tables in csv?

[eric-czech]

Ah, I didn't know their cross-filtering was broken.

I would say that the current cytokit explorer does do that since it only needs a csv with object features, object images (i.e. cell boundaries) and original images. If you created your own object images and feature tables, it wouldn't be too hard to conform that data with what Cytokit expects. This would involve renaming csv fields to match expectations as well as ensuring that the labeled images are in 5D tiff tiles for each field of view within a grid. You would also need to either use Cytokit or something else to create the downsampled stitched image explorer uses for tile navigation. If you want to try to go that route, I'd happily lend some advice/support as you go through it, but I suspect it will be difficult without digging in to the source code a bit.