Is there a way to run k-means only clustering inside scimap module?

[emir-radkevich]

Hi.
I would like to have k-means and leiden clustering outputs only. Is there a way to specify that behaviour in params.yml?
Thanks.

[ArtemSokolov]

Hi @emir-radkevich,

Looking at the code, it doesn't look like it. Seems the wrapper purposefully runs all four clustering methods.

@ajitjohnson: Would it be possible to expose --clustering-method to the CLI user?

[ajitjohnson]

@ArtemSokolov

exposed a parameter called method

# single method
python -m scimap.cli._scimap_mcmicro input.csv -o output_dir --method kmeans

# multiple methods
python -m scimap.cli._scimap_mcmicro input.csv -o output_dir --method kmeans leiden

# all methods
python -m scimap.cli._scimap_mcmicro input.csv -o output_dir --method all

let me know if it works. I can then rebuild the docker for mcmicro.

[ArtemSokolov]

Thanks so much, @ajitjohnson.

I would say let's go ahead and build the container. We will want to test it in a containerized workflow anyway.

Based on this CI script, I think it will automatically docker build and docker push when you make a new release.

[emir-radkevich]

Thanks.

As a follow-up question, what is the difference between scanpy and scimap steps?
I thought scanpy calculates leiden clustering, scimap -- k-means, phenograph and spatial. Based on the code, leiden is calculated in scimap as well.

[ArtemSokolov]

They are just different toolboxes developed by different people.

Scimap exposes clustering to MCMICRO, but the toolbox itself is standalone and has a lot of other functionality for doing downstream analysis of spatial data: https://scimap.xyz/

"scanpy" is one of three clustering modules that were developed to have standardized i/o interface, with the other two being:

• https://github.com/labsyspharm/mcmicro-fastPG
• https://github.com/labsyspharm/mcmicro-flowSOM
  The ultimate goal of these were to feed a consensus clustering method, though it never quite made it as a formal step in MCMICRO.