Add preprint use case tutorials #48
Conversation
|
Check out this pull request on See visual diffs & provide feedback on Jupyter Notebooks. Powered by ReviewNB |
|
View / edit / reply to this conversation on ReviewNB Zethson commented on 2024-09-27T15:22:38Z Can you add 1 more sentence of context of what the goal of this analysis is and why it matters, please? Lilly-May commented on 2024-09-29T10:01:05Z I added: Overall, this tutorial demonstrates how to use pertpy to derive insights into perturbation responses in a complex dataset comprising various cell lines and drugs, with the goal of better understanding the molecular mechanisms underlying drug responses and how they vary across cell lines. |
|
View / edit / reply to this conversation on ReviewNB Zethson commented on 2024-09-27T15:22:39Z Should we show obs here once? And maybe describe the types of perturbations? Lilly-May commented on 2024-09-29T10:05:33Z Done! |
|
View / edit / reply to this conversation on ReviewNB Zethson commented on 2024-09-27T15:22:40Z Line #6. #sc.pp.highly_variable_genes(adata, subset=True, n_top_genes=4000) Why is it documented out? Either it was run or not. Lilly-May commented on 2024-09-29T10:06:21Z Yes, true - I deleted it. |
|
View / edit / reply to this conversation on ReviewNB Zethson commented on 2024-09-27T15:22:41Z 1 sentence of motivation why this is useful and we're doing it would be awesome. Lilly-May commented on 2024-09-29T10:10:45Z I changed the paragraph to: Datasets often come with metadata that can enable more detailed analyses, as those presented in this tutorial. However, the extent of metadata can vary greatly between datasets, and each dataset usually has its own metadata format. To address this, databases that allow for standardized metadata annotation exist. Pertpy offers multiple metadata classes to query these databases and annotate your data. Here, we will use the |
|
View / edit / reply to this conversation on ReviewNB Zethson commented on 2024-09-27T15:22:42Z Ensure that It is, right? Is this a misleading message? Lilly-May commented on 2024-09-29T10:15:46Z Yes, if the Zethson commented on 2024-09-29T10:21:09Z Yeah so removing the line breaks helps. But I think that the message "ensure that it's available" should not written in the first place. The code should just check whether this column is inside the object and if not -> error with a message. Else, don't print this useless logging message.
In general, the UX is suboptimal here. There should simply be better inference of parameters and just defaults. Lilly-May commented on 2024-09-29T13:06:44Z I opened an issue (scverse/pertpy#664) because I think this requires a bit more discussion and possibly a general revision of the metadata UX. |
|
View / edit / reply to this conversation on ReviewNB Zethson commented on 2024-09-27T15:22:43Z This might require one comment of justification? Maybe not necessary idk. Up to you. Lilly-May commented on 2024-09-29T10:17:58Z I've added one. |
|
View / edit / reply to this conversation on ReviewNB Zethson commented on 2024-09-27T15:22:44Z Ideally we wouldn't tell the users to "ensure" but rather we check and if it's not the case, we print a warning. Can you either fix this or open an issue, please? Lilly-May commented on 2024-09-29T11:21:55Z This is what is done - here the respective section in the code.
if sum(adata.obsm[metadata_key].columns != adata.var.index.values) > 0:
logger.warning(
"Column name of metadata is not the same as the index of adata.var. Ensure that the genes are in the same order."
)
So, we could either just change the warning and make the message clearer or we raise an Error and let the method fail when this happens? Zethson commented on 2024-09-29T11:23:38Z Yeah if it's critical for the usage of the function it should error. Else we should clarify the effect that this has. If it has no effect at all -> remove warning. Lilly-May commented on 2024-09-29T13:06:56Z I opened an issue (scverse/pertpy#664) for this because I think this requires a bit more discussion and possibly a general revision of the metadata UX. |
|
View / edit / reply to this conversation on ReviewNB Zethson commented on 2024-09-27T15:22:45Z Is there a way for us to use the pertpy DE volcano plot here? I know that we might need something tuned for the paper... Lilly-May commented on 2024-09-29T11:45:16Z Yes, we can - thanks for noting that! I changed it. |
|
View / edit / reply to this conversation on ReviewNB Zethson commented on 2024-09-27T15:22:46Z 1 sentence on what this analysis is biologically about or why it matters, please. Lilly-May commented on 2024-09-29T12:07:33Z I changed the sentence to: We will then use the identified perturbed cells to compute a perturbation space and explore the genetic interaction manifolds, meaning we will explore how distinct perturbations target similar or different molecular mechanisms and how these effects can be visualized in a low-dimensional space. |
|
View / edit / reply to this conversation on ReviewNB Zethson commented on 2024-09-27T15:22:47Z 1 sentence on the interpretation maybe? I know that it got criticized but it's in the paper as well. Fine for now. Lilly-May commented on 2024-09-29T11:57:15Z I added: Pertpy facilitates the preprocessing of perturbation datasets, for example, by removing confounding effects using Mixscape. We can then use the identified perturbed cells to compute a perturbation space and explore genetic interaction manifolds. Here, we successfully reproduced the results presented by Norman et al. (2019), as perturbations with the same pre-annotated gene programs cluster together in the pseudobulked perturbation space. Additionally, we extended the gene program annotation to perturbations with unannotated gene programs using label transfer. Overall, this analysis enables the exploration of how distinct perturbations target similar or different molecular mechanisms. |
|
View / edit / reply to this conversation on ReviewNB Zethson commented on 2024-09-27T15:22:48Z Line #1. def filter_data(adata_temp): Maybe rename this to something more descriptive? Lilly-May commented on 2024-09-29T12:09:22Z I renamed it to |
|
View / edit / reply to this conversation on ReviewNB Zethson commented on 2024-09-27T15:22:49Z Can we add at least 1 sentence of takeaway here? What do the plots mean? Lilly-May commented on 2024-09-29T12:41:22Z I added: The heatmaps above show that patients who responded to chemotherapy showed a larger difference between their pre- and post-treatment expression profiles compared to those who responded to the combination of anti-PDL-1 and chemotherapy. This indicates that the combination therapy may have led to a less intense response or was applied in cases with poorer prognoses. |
|
View / edit / reply to this conversation on ReviewNB Zethson commented on 2024-09-27T15:22:50Z Line #1. # Get reference cell type That's a comment for line 13 no? Lilly-May commented on 2024-09-29T12:00:55Z The entire code cell is just there to obtain the reference cell type for the entire dataset, i.e. the adata which includes all treatments. I then create new Sccoda models for the adatas with individual treatments. |
|
View / edit / reply to this conversation on ReviewNB Zethson commented on 2024-09-27T15:22:51Z Can we get a bit more interpretation here? What do the effects mean etc? Just some bla from the paper or so is good enough. Lilly-May commented on 2024-09-29T12:45:06Z I added: This fits to our earlier findings from the distance metric analysis, where we observed that the Anti-PD-L1 + Chemo group showed a smaller difference between their pre- and post-treatment expression profiles compared to the Chemo group. Overall, these findings indicate that the Anti-PD-L1 + Chemo combination therapy may have led to a less intense response or was applied in cases with poorer prognoses. |
|
I added: Overall, this tutorial demonstrates how to use pertpy to derive insights into perturbation responses in a complex dataset comprising various cell lines and drugs, with the goal of better understanding the molecular mechanisms underlying drug responses and how they vary across cell lines. View entire conversation on ReviewNB |
|
Done! View entire conversation on ReviewNB |
|
Yes, true - I deleted it. View entire conversation on ReviewNB |
|
I changed the paragraph to: Datasets often come with metadata that can enable more detailed analyses, as those presented in this tutorial. However, the extent of metadata can vary greatly between datasets, and each dataset usually has its own metadata format. To address this, databases that allow for standardized metadata annotation exist. Pertpy offers multiple metadata classes to query these databases and annotate your data. Here, we will use the View entire conversation on ReviewNB |
|
Yes, if the View entire conversation on ReviewNB |
|
I've added one. View entire conversation on ReviewNB |
|
Yeah so removing the line breaks helps. But I think that the message "ensure that it's available" should not written in the first place. The code should just check whether this column is inside the object and if not -> error with a message. Else, don't print this useless logging message.
In general, the UX is suboptimal here. There should simply be better inference of parameters and just defaults. View entire conversation on ReviewNB |
|
This is what is done - here the respective section in the code.
if sum(adata.obsm[metadata_key].columns != adata.var.index.values) > 0:
logger.warning(
"Column name of metadata is not the same as the index of adata.var. Ensure that the genes are in the same order."
)
So, we could either just change the warning and make the message clearer or we raise an Error and let the method fail when this happens? View entire conversation on ReviewNB |
|
Yeah if it's critical for the usage of the function it should error. Else we should clarify the effect that this has. If it has no effect at all -> remove warning. View entire conversation on ReviewNB |
|
Yes, we can - thanks for noting that! I changed it. View entire conversation on ReviewNB |
|
I added: Pertpy facilitates the preprocessing of perturbation datasets, for example, by removing confounding effects using Mixscape. We can then use the identified perturbed cells to compute a perturbation space and explore genetic interaction manifolds. Here, we successfully reproduced the results presented by Norman et al. (2019), as perturbations with the same pre-annotated gene programs cluster together in the pseudobulked perturbation space. Additionally, we extended the gene program annotation to perturbations with unannotated gene programs using label transfer. Overall, this analysis enables the exploration of how distinct perturbations target similar or different molecular mechanisms. View entire conversation on ReviewNB |
|
The entire code cell is just there to obtain the reference cell type for the entire dataset, i.e. the adata which includes all treatments. I then create new Sccoda models for the adatas with individual treatments. View entire conversation on ReviewNB |
|
I changed the sentence to: We will then use the identified perturbed cells to compute a perturbation space and explore the genetic interaction manifolds, meaning we will explore how distinct perturbations target similar or different molecular mechanisms and how these effects can be visualized in a low-dimensional space. View entire conversation on ReviewNB |
|
I renamed it to View entire conversation on ReviewNB |
|
I added: The heatmaps above show that patients who responded to chemotherapy showed a larger difference between their pre- and post-treatment expression profiles compared to those who responded to the combination of anti-PDL-1 and chemotherapy. This indicates that the combination therapy may have led to a less intense response or was applied in cases with poorer prognoses. View entire conversation on ReviewNB |
|
I added: This fits to our earlier findings from the distance metric analysis, where we observed that the Anti-PD-L1 + Chemo group showed a smaller difference between their pre- and post-treatment expression profiles compared to the Chemo group. Overall, these findings indicate that the Anti-PD-L1 + Chemo combination therapy may have led to a less intense response or was applied in cases with poorer prognoses. View entire conversation on ReviewNB |
|
I opened an issue (scverse/pertpy#664) because I think this requires a bit more discussion and possibly a general revision of the metadata UX. View entire conversation on ReviewNB |
|
I opened an issue (scverse/pertpy#664) for this because I think this requires a bit more discussion and possibly a general revision of the metadata UX. View entire conversation on ReviewNB |
|
Thanks! Feel free to merge and pull then. |
This PR adds the three use cases from the pertpy preprint to the tutorials. I created a separate notebook for each of the Norman, McFarland, and Zhang use cases. For the Zhang use case, the Dialogue section is still missing. I’ll add that once this PR is merged, as I need to run it on the cluster.