Add new method: CellMapper

[Marius1311]

Describe your changes

This adds CellMapper as a new denoising method, see https://github.com/quadbio/cellmapper

Checklist before requesting a review

• I have performed a self-review of my code

• Check the correct box. Does this PR contain:

  • Breaking changes
  • New functionality
  • Major changes
  • Minor changes
  • Bug fixes

• Proposed changes are described in the CHANGELOG.md

• CI Tests succeed and look good!

[LuckyMD]

Looks good code-wise from my side, but it's unclear to me how this compares to knn-smoothing.

Would need a review from @wes-lewis and/or @rcannood or @scottgigante to confirm though, as I'm not as familiar with the structure in this task.

[wes-lewis]

[To address what @LuckyMD mentioned] This is quite different from the already implemented knn_smoothing approach. knn_smoothing sounds as though it would be a very general approach, but the original code from Yanai Lab uses quite specific methodology. Considering this, it should be different enough to be included. I'll describe differences a bit, as they are more specific than what is covered in the viash description (may be useful to update the description to ensure differences are accurately reflected)

knn_smoothing specifics:

• Normalizations: median normalization, Freeman Tukey transform)
• Iteratively increasing neighborhood size: Progressively increasing k_step up to k = k
• Iterative neighborhood updates: In each iteration, a new matrix smoothed matrix X is calculated where the expression vector for a given cell i is replaced by the sum of the raw (unsmoothed) expression values over all of its neighbors (neighbors calculated via normalizing and calling euclidian distance on either the smoothed matrix from the previous iteration, or the raw/unsmoothed expression values on the first iteration). This summation is not weighted (e.g. by the euclidian distances or a kernel), which differs from CellMapper.
• Output: After summing unnormalized expression values over neighbors in the final step, there is no integrated normalization step (resulting output is relatively inflated, by on average k*X[:,i] )

These aspects all differ from the choices used in CellMapper, which continues to update expression values iteratively and weights neighborhood members according to a user-defined kernel. CellMapper smoothing is more analogous to diffusion methods and less to the knn_smooth approach.

I do not see any issue with the code, other than the few typos pointed our by me and @LuckyMD. I'm happy to merge as soon as these are fixed.

[Marius1311]

Thanks for the review @LuckyMD @wes-lewis , I fixed the typos and extended the description in the config file, should be ready now.

[Marius1311]

FYI I added myself as a contributor in the _viash.yaml

[rcannood]

LGTM as far as I'm concerned! @wes-lewis @LuckyMD What about you?