Add reworked TemporalProblem tutorial based on pancreas dataset

[LeonStadelmann]

No description provided.

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[review-notebook-app]

View / edit / reply to this conversation on ReviewNB

MUCDK commented on 2024-09-20T04:44:23Z
----------------------------------------------------------------

nice, please say that "X_MultiVI" was generated with "MultiVI", and link to the page https://docs.scvi-tools.org/en/stable/user_guide/models/multivi.html

[review-notebook-app]

View / edit / reply to this conversation on ReviewNB

MUCDK commented on 2024-09-20T04:44:24Z
----------------------------------------------------------------

"As our goal is to study endocrine cells, we filter out ductal and acinar cells. Filtering to the relevant cell types often helps to get a more accurate mapping"

[review-notebook-app]

View / edit / reply to this conversation on ReviewNB

MUCDK commented on 2024-09-20T04:44:25Z
----------------------------------------------------------------

"As the TemporalProblem (obviously with link) requires the temporal information to be numeric, we create a column ... of numeric type."

[review-notebook-app]

View / edit / reply to this conversation on ReviewNB

MUCDK commented on 2024-09-20T04:44:26Z
----------------------------------------------------------------

I might have done it in a different way in the analysis notebooks, but let's start here with initializing the model and preparing the model . The computation of the cost matrices would then go below "prepare", as it is only required for "solve'

[review-notebook-app]

View / edit / reply to this conversation on ReviewNB

MUCDK commented on 2024-09-20T04:44:26Z
----------------------------------------------------------------

Great. Now you can say that up to here, moscot would use the default cost function, which is sq. eucl. distance.

Instead, we want to use geodesic distances, for which we need to create graphs. Then you can insert the code chunk to create the graphs

[review-notebook-app]

View / edit / reply to this conversation on ReviewNB

MUCDK commented on 2024-09-20T04:44:27Z
----------------------------------------------------------------

Looks good, thanks for that.

For the next iteration, let's try to add the part about how to identify driver genes / TFs as done here: https://github.com/theislab/moscot-framework_reproducibility/blob/main/notebooks/time/multi-modality/16_OT_delta_eps_analysis/04_marker_features.ipynb

I.e., we want to have the cells 20-32 where we identify Neurod2 to be the second highest TF for eps. prog. and Fev+ Delta

[review-notebook-app]

View / edit / reply to this conversation on ReviewNB

MUCDK commented on 2024-10-08T06:36:38Z
----------------------------------------------------------------

Can we use the threshold argument to only display transition with at least 0.05? If it doesn't help, maybe try decreasing the font size? should be possible , at least via kwargs,but check out the docs, and if not documented, in the code.

[review-notebook-app]

View / edit / reply to this conversation on ReviewNB

MUCDK commented on 2024-10-08T06:36:39Z
----------------------------------------------------------------

let's remove this (also for Fev+ delta), and only keep the transcription factor analysis (i.e. where you set features="mouse")

[review-notebook-app]

View / edit / reply to this conversation on ReviewNB

MUCDK commented on 2024-10-08T06:36:40Z
----------------------------------------------------------------

In the next cell, add an overall comment along the lines of: "Our analysis suggests that Neurod2 is a potential lineage driver for both epsilon and delta cells. As described in the manuscript (link to the manuscirpt), we thus functionally validated this hypothesis and found Neurod2 to be indeed an activator of epsilon cell regulation."

[MUCDK]

hmm also the driver feature analysis reveals similar results, but not exactly the same ones. I will think about how to address that. We might have to look into the two different versions, and see where the discrepancy comes from.

[review-notebook-app]

View / edit / reply to this conversation on ReviewNB

MUCDK commented on 2024-10-17T09:00:34Z
----------------------------------------------------------------

Did you check that the entries are like in the original notebook?

[review-notebook-app]

View / edit / reply to this conversation on ReviewNB

MUCDK commented on 2024-10-17T09:00:35Z
----------------------------------------------------------------

also here just please confirm that it's the same values.

[review-notebook-app]

View / edit / reply to this conversation on ReviewNB

MUCDK commented on 2024-10-17T09:00:36Z
----------------------------------------------------------------

Can we do ancestors, and not descendants here , please?

[review-notebook-app]

View / edit / reply to this conversation on ReviewNB

MUCDK commented on 2024-10-17T09:00:37Z
----------------------------------------------------------------

it's not a "be a highly correlated transcription factor in Fev+ delta cells", but the "predicted progenitor population of Fev+ delta cells specifically expresses Neurod2".

Same for epsilon cells.

Let's rephrase the last sentence (as atm you're talking about Fev+ delta cells, but then conclude on Epsilon cells): "Thus, we found a potential transcription factor for both delta and epsilon populations. Thus, we set out to experimantally verify the role of Neurod2 (italic), and found it to be an activator of epsilon cells. (manuscirpt)".