DEGA-106-Segmentation-Metrics

[cornhundred]

Adding a qc module that will contain a iST segmentation metrics method.

[huanlity]

Hi @jaspreetishar, great work! I had some minor comments that is focusing on readability, simplicity, and reproducibility of our codebase. Other than that, can share the data dir data/segmentation_metrics_data so I can do a quick test of the QC notebook? Thanks!

[huanlity]

As long as you change this line in the notebook:

directories = [f"{dataset_name.lower()}_{d.lower().replace('-','_')}" for d in segmentation_approaches]

to this to prevent a file existence error:

directories = [
    f"{dataset_name.lower().replace('-','_')}_{d.lower().replace('-','_')}" for d in segmentation_approaches
    ]

I tested, everything LGTM! Great job! ✅

[jaspreetishar]

As long as you change this line in the notebook:

directories = [f"{dataset_name.lower()}_{d.lower().replace('-','_')}" for d in segmentation_approaches]

to this to prevent a file existence error:

directories = [
    f"{dataset_name.lower().replace('-','_')}_{d.lower().replace('-','_')}" for d in segmentation_approaches
    ]

I tested, everything LGTM! Great job! ✅

I have made the changes - thanks so much for the helpful review!!

[jaspreetishar]

Hi @cornhundred, the branch has been reviewed. If everything looks good on your end as well, could we go ahead and merge it? Thanks!

[cornhundred]

Possibly rename "mixed_expression_calc" to something like "orthogonal_expression_calc". Also, add quantification to this method. Calculate proportion of cell-type-A cells that express cell-type-B genes above some threshold and vice versa. We can also sanity check that a cell expressing an orthogonal marker from a different cell type is actually neighboring a cell from the orthogonal cell type.

[cornhundred]

@jaspreetishar we can drop the subset_interval_y_x from the docstring

[cornhundred]

I do not think we need to transform transcripts or cell polygons to image coordinates for segmentation QC. We also need to have a version of the STP segmentation data in micron coordinates.