Mixing score

[camisowers]

If you haven't already, please read through our contributing guidelines before opening your PR

What is the purpose of this PR?
Closes #756 and closes #793.
Closes #906.

How did you implement your changes
Detailed in design doc.
In addition to adding cell cluster column name to neighborhood matrix, I also added a mixing arg (default False) which indicates whether to drop cells with no nearby neighbors from the table.
(It was doing this automatically, but we would want to keep these cells in the table to determine if FOVs are cold by summing the reference cells.)

• edit: Additionally, we will ensure that any cells not included in cluster analysis are still included in the resulting cell table, as noted in Fix cell neighbors matrix/clustering dropping for cells that don't have any neighbors #906.

Remaining issues

• test against Leeat's data
• test against Noah's data
• check cells with no cluster assignment stay in table
• check 50 vs 100 radius
• check count vs frequency

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[camisowers]

Testing against Leeat's data and using the calculation detailed here, we see our results below on the left. There only are two misclassifications seen; however while we visualize the raw scores, her data seems to have some type of normalizing, but it is not mentioned in her paper.

If we were to consider a "percent mixing score" which I noticed she explored in her code, we see a more similar representation of the mixing scores but with now an additional misclassification. All three are borderline cases though and don't raise too much concern.

Differences between the two calculations:

Adjusting the radius for creating neighborhood matrices seems to slightly change the order of the FOV mixing scores, but there doesn't seem to be any drastic changes. This would mostly affect the borderline cases and determining a threshold.

[ngreenwald]

Looks great, see comments on the notebook

[camisowers]

Notebook now generates a simple plot

[ngreenwald]

Looks good! Now that we've worked out the logic for dropping rare cells, same question as before: do we need to separate the different spatial analyses?

[ngreenwald]

Everything else looks good