fixes dendrogram issue #1228

[fidelram]

The PR uses 1-correlation as distance matrix to compute the dendrogram as suggested in #1288 and mentioned in scverse/squidpy#236

I opted for the minimal changes to the code. Other solution would be to use scipy.spatial.distance.pdist that allows a larger number of distance metrics. I am open for a discussion on this topic (ping @michalk8)

[giovp]

Hi @fidelram !
looks good yes, with distance.squareform you are essentially transforming the corr in a condensed distance matrix right? and that's what pdist also passes. So I think it looks good now!
Another option would have been to pass mean_df.T directly to linkage, but I understand wanting to keep correlation as distances.

[fidelram]

Hi @giovp, basically I mimic the what the linkage code would do internally when the first argument is not a distance. Because the correlation matrix is used for sc.tl.correlation_matrix it is kept.

[giovp]

Makes sense! Thanks it's clear now why the corr matrix

[ivirshup]

Could this get:

• Bugfix release note
• Links to issues that should be closed by this

[fidelram]

Issues that should be closed:

• Probably wrong distance matrix used in sc.tl.dendrogram #1288

[ivirshup]

@meeseeksdev backport to 1.7.x

[ivirshup]

@fidelram, in future, if you add a line like:

Fixes #x, #y

To the initial comment in a PR, those issues will automatically be closed when the PR is merged. I think you can also link PRs to issues with the "Linked issues" element in the righthand sidebar. Using these makes it easier to tell which issues have been addressed already.