Update kmeans neighborhood notebook take 2 #714

cliu72 · 2022-09-17T00:07:55Z

What is the purpose of this PR?

Update and clean-up the k-means neighborhood notebook.

How did you implement your changes

Fixed a bug where the wrong cell phenotypes were being counted due to fovs missing certain cell phenotypes.
The old notebook required the cell table to include both a column that specified a numeric id for each cell phenotype, and a column that had the cell phenotypes. I simplified this to only require the column with the cell phenotypes. Specifically, in create_neighborhood_matrix, removed cluster_id_col, which was redundant with cluster_name_col.
Set default of self_neighbor to False, so a cell is not counting itself as its neighbor.
Set cluster numbers to start at 1. Previously was starting at 0, which is confusing because in cell overlays, 0 usually corresponds to empty slide.
Add inertia as a metric.
Subsample cells for silhouette score calculation.
Remove cells that have no neighbors before clustering.
Scaled (z-scored) the heatmaps, which makes them more interpretable.
Added heatmap that shows the composition of each neighborhood (basically the average feature vector for each cluster).
Fixed overlay generation function so correct legend is shown.
Added subheadings and more descriptive text to notebook.

Remaining issues

Need to talk about whether to delete the part of the spatial LDA notebook that does kmeans

review-notebook-app · 2022-09-17T00:07:58Z

Check out this pull request on

See visual diffs & provide feedback on Jupyter Notebooks.

Powered by ReviewNB

cliu72 · 2022-09-19T19:27:18Z

@ngreenwald wtf

templates_ark/example_neighborhood_analysis_script.ipynb

ngreenwald · 2022-09-20T00:37:26Z

templates_ark/example_neighborhood_analysis_script.ipynb

@@ -1,18 +1,32 @@
 {


See if you can use a smaller heading size for this text.

Reply via ReviewNB

I'm not sure why it's showing up as huge in ReviewNB. It looks fine on github (https://github.com/angelolab/ark-analysis/blob/kmeans_neighborhood_nb_update/templates_ark/example_neighborhood_analysis_script.ipynb) and looks fine in my jupyter.

ngreenwald · 2022-09-20T00:37:26Z

templates_ark/example_neighborhood_analysis_script.ipynb

@@ -1,18 +1,32 @@
 {


What about adding a min_k as well? Did you ever look at k values below 5?

Reply via ReviewNB

Yes, I have looked at k values below 5 when looking at specific regions of the lymph node (i.e. what are the neighborhoods only within the T cell zone?). But I added a min_k anyways for flexibility.

templates_ark/example_neighborhood_analysis_script.ipynb

ngreenwald · 2022-09-20T00:37:27Z

templates_ark/example_neighborhood_analysis_script.ipynb

@@ -1,18 +1,32 @@
 {


Grande text size again

Reply via ReviewNB

Same as above, it looks fine on github (https://github.com/angelolab/ark-analysis/blob/kmeans_neighborhood_nb_update/templates_ark/example_neighborhood_analysis_script.ipynb) and looks fine in my jupyter.

ngreenwald · 2022-09-20T00:37:27Z

templates_ark/example_neighborhood_analysis_script.ipynb

@@ -1,18 +1,32 @@
 {


Is the goal to replace all of this with Mantis once we have the single directory saving implemented? Or some of these plots should be kept even once we make that change?

Reply via ReviewNB

If I understand it correctly, the mantis function (create_mantis_dir) just copies all the masks over, so the single directory saving will just take these masks and copy them over. This function actually creates the overlays.

ngreenwald

Looking great, some additional small comments here

ark/analysis/spatial_analysis.py

ngreenwald · 2022-09-20T00:48:28Z

ark/analysis/spatial_analysis_test.py

-    assert (counts.loc[:9, "Pheno2"] == 8).all()
-    assert (counts.loc[10:19, "Pheno3"] == 8).all()
+    # test the counts values
+    assert (counts[(counts[settings.FOV_ID] == "fov8") &


Were the tests not using two FOVs before? Why are we now indexing for specific FOVs, and specific segmentation labels within those FOVs?

I honestly don't know what was going on with the tests before. After fixing the bug in the code, those tests were just straight up wrong. I manually confirmed that this version is correct (literally just counted the values in the data table). Feel free to check my work.

Yeah, sometimes ReviewNB does weird stuff. Once you've addressed all of the comments hit the re-request review button at the top, that way I know you're actually finished, easier for you than responding to each comment individually.

ngreenwald · 2022-09-20T00:52:27Z

ark/utils/spatial_analysis_utils.py


    # dot binarized 'is neighbor?' matrix with pheno_has_cell to get counts
    counts = pheno_has_cell.dot(cell_dist_mat_bin).T
+    counts_pd = pd.DataFrame(counts, columns=pheno_names_from_tab)
+    counts_pd = counts_pd.set_index(current_fov_neighborhood_data.index.copy())


Is this for code readability or is fixing a bug?

Fixing a bug. Setting the columns is important - not all FOVs have the same cell types. This was the problem before that was leading to the wrong results. Setting the index is needed to merge the tables for all FOVs.

ark/analysis/spatial_analysis_test.py

ark/utils/spatial_analysis_utils.py

ark/utils/spatial_analysis_utils_test.py

…gelolab/ark-analysis into kmeans_neighborhood_nb_update

alex-l-kong

Made some additional comments, mostly regarding the notebook.

ark/utils/spatial_analysis_utils.py

templates_ark/example_neighborhood_analysis_script.ipynb

…gelolab/ark-analysis into kmeans_neighborhood_nb_update

templates_ark/example_neighborhood_analysis_script.ipynb

alex-l-kong · 2022-09-23T19:24:38Z

templates_ark/example_neighborhood_analysis_script.ipynb

@@ -1,18 +1,34 @@
 {


Is there any way we can adjust the colorbar width so it doesn't get too close to the heatmap?

Reply via ReviewNB

This is an issue with all places where visualize.draw_heatmap is called. I noticed it when testing the cell clustering notebook. I believe it is also used in the spatial enrichment notebook. I think we can leave this for another pull request, so whoever makes the fix can confirm the heatmaps look okay in all of those notebooks.

I also noticed just now that some of the other example notebooks do not have example output visualizations, as you have requested for this one. I think the pixel clustering and cell clustering notebooks, and spatial enrichment notebooks would also benefit from example outputs.

@cliu72 seems like the sample visualizations I had in the pixel and cell clustering notebooks got removed during some recent PRs, I'll ask @srivarra put those back in while he addresses the example dataset PR on his end.

This is an issue with all places where visualize.draw_heatmap is called. I noticed it when testing the cell clustering notebook. I believe it is also used in the spatial enrichment notebook. I think we can leave this for another pull request, so whoever makes the fix can confirm the heatmaps look okay in all of those notebooks.

Yeah, the underlying heatmap function unfortunately doesn't offer great documentation in order to fix this. I think it'll require looking under the hood to see what internal properties are being set, I had to do that with a different seaborn function somewhere.

@cliu72 @alex-l-kong Are the sample visualizations the plots generated by the notebook? So the notebooks should be saved with their output contained from each cell?

@srivarra yes, the overlays for both notebooks as well as the weighted channel heatmaps for the cell clustering notebook.

Ok, Sounds good!

alex-l-kong

Looks good, just a couple more minor comments on the notebook.

alex-l-kong

Looks good to me.

cliu72 · 2022-09-26T21:59:10Z

Addressed all comments. @ngreenwald @ackagel anything to add?

ngreenwald

Looks good!

cliu72 added 12 commits August 24, 2022 03:48

Update kmeans neighborhood notebook

35acc6a

Revert start_docker.sh tags

35c3cba

Revert example data

c78cf3c

merged main

bf27f1b

Revert example data

86c4f4c

Update neighborhood code

8024d8c

pycodestyle fix

5530d1d

Revert settings.py to avoid problems with tests

06bbf71

Fix notebook test and remove mantis for now

68fbf4e

Fix tests

806855e

Fix silhouette score function description

326830f

Fix typo in notebook

930d928

cliu72 requested a review from ngreenwald September 17, 2022 00:38

ngreenwald reviewed Sep 20, 2022

View reviewed changes

ngreenwald requested changes Sep 20, 2022

View reviewed changes

ngreenwald requested review from ackagel and alex-l-kong September 20, 2022 00:53

cliu72 added 5 commits September 20, 2022 02:58

Add min_k

38423ab

Fix add 1 to generate_cluster_labels

9c7bffd

Breakup lines that removes cells with no neighbors

99b8669

Update notebook

1e820d8

Merge branch 'main' into kmeans_neighborhood_nb_update

f51e557

cliu72 requested a review from ngreenwald September 20, 2022 17:03

alex-l-kong reviewed Sep 21, 2022

View reviewed changes

ark/analysis/spatial_analysis_test.py Show resolved Hide resolved

ark/utils/spatial_analysis_utils.py Outdated Show resolved Hide resolved

ark/utils/spatial_analysis_utils_test.py Show resolved Hide resolved

cliu72 added 4 commits September 21, 2022 20:46

Change silhouette function code

65a58ec

Add check for min_k

62db01f

Add tests

5b4c241

Merge branch 'kmeans_neighborhood_nb_update' of https://github.com/an…

dd833ca

…gelolab/ark-analysis into kmeans_neighborhood_nb_update

Merge branch 'main' into kmeans_neighborhood_nb_update

5b3f568

cliu72 requested a review from alex-l-kong September 21, 2022 21:42

alex-l-kong requested changes Sep 22, 2022

View reviewed changes

ark/utils/spatial_analysis_utils.py Outdated Show resolved Hide resolved

alex-l-kong reviewed Sep 22, 2022

View reviewed changes

templates_ark/example_neighborhood_analysis_script.ipynb Show resolved Hide resolved

templates_ark/example_neighborhood_analysis_script.ipynb Show resolved Hide resolved

cliu72 added 4 commits September 22, 2022 18:49

Change cluster label code

dba7407

Add notebook plots

c6bd77b

Merge branch 'kmeans_neighborhood_nb_update' of https://github.com/an…

13141a7

…gelolab/ark-analysis into kmeans_neighborhood_nb_update

Merge branch 'main' into kmeans_neighborhood_nb_update

9871dec

cliu72 requested a review from alex-l-kong September 23, 2022 18:03

alex-l-kong reviewed Sep 23, 2022

View reviewed changes

Change error message to ValueError

067fc9b

cliu72 requested a review from alex-l-kong September 24, 2022 01:43

alex-l-kong reviewed Sep 26, 2022

View reviewed changes

srivarra added the enhancement New feature or request label Sep 26, 2022

ngreenwald approved these changes Sep 26, 2022

View reviewed changes

ngreenwald merged commit d4a246a into main Sep 26, 2022

ngreenwald deleted the kmeans_neighborhood_nb_update branch September 26, 2022 23:13

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Update kmeans neighborhood notebook take 2 #714

Update kmeans neighborhood notebook take 2 #714

cliu72 commented Sep 17, 2022

review-notebook-app bot commented Sep 17, 2022

cliu72 commented Sep 19, 2022

ngreenwald Sep 20, 2022 •

edited

Loading

cliu72 Sep 20, 2022

ngreenwald Sep 20, 2022

cliu72 Sep 20, 2022 •

edited

Loading

ngreenwald Sep 20, 2022 •

edited

Loading

cliu72 Sep 20, 2022

ngreenwald Sep 20, 2022

cliu72 Sep 20, 2022

ngreenwald left a comment

ngreenwald Sep 20, 2022

cliu72 Sep 20, 2022

ngreenwald Sep 20, 2022

ngreenwald Sep 20, 2022

cliu72 Sep 20, 2022

alex-l-kong left a comment

alex-l-kong Sep 23, 2022

cliu72 Sep 24, 2022

cliu72 Sep 24, 2022

alex-l-kong Sep 26, 2022

alex-l-kong Sep 26, 2022

srivarra Sep 26, 2022

alex-l-kong Sep 26, 2022

srivarra Sep 26, 2022

alex-l-kong left a comment

alex-l-kong left a comment

cliu72 commented Sep 26, 2022

ngreenwald left a comment

Update kmeans neighborhood notebook take 2 #714

Update kmeans neighborhood notebook take 2 #714

Conversation

cliu72 commented Sep 17, 2022

review-notebook-app bot commented Sep 17, 2022

cliu72 commented Sep 19, 2022

ngreenwald Sep 20, 2022 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

cliu72 Sep 20, 2022 • edited Loading

Choose a reason for hiding this comment

ngreenwald Sep 20, 2022 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ngreenwald left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

alex-l-kong left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

alex-l-kong left a comment

Choose a reason for hiding this comment

alex-l-kong left a comment

Choose a reason for hiding this comment

cliu72 commented Sep 26, 2022

ngreenwald left a comment

Choose a reason for hiding this comment

ngreenwald Sep 20, 2022 •

edited

Loading

cliu72 Sep 20, 2022 •

edited

Loading

ngreenwald Sep 20, 2022 •

edited

Loading