Add remaining text to BANKSY notebook#1010
Conversation
jashapiro
left a comment
jashapiro
left a comment
There was a problem hiding this comment.
Overall this looks good; most of my comments are small, and the ones that are a bit larger you probably knew to expect... care about over-interpretation and a bias toward questions not conclustions.
Other than that, my biggest "major edit" is consistency in how we style "BANKSY" when we are not referring to code. I would say we should not put it in code style most of the time when we referring to the approach/algorithm, not the package. When we do refer to the package, we should probably be styling it Banksy, as that is how it appears in code.
| For example, we can see from the coordinate plots that non-spatial clusters 3 and 5 correspond to spots likely dominated by tumor cells, which has been split into several BANKSY AGF clusters including at least 3, 5, and 7. | ||
| The heatmap helps us see how non-spatial clusters map onto the finer AGF cluster structure. | ||
| For example, we can see from the coordinate plots that non-spatial clusters 3 and 5 contain spots that are likely dominated by tumor cells. | ||
| These clusters have been split into several `BANKSY` AGF clusters, suggesting that AGF is resolving finer spatial structure within regions that expression-only clustering treated as a single domain. |
There was a problem hiding this comment.
How do you know this is spatial structure and not over-clustering? I would be cautious with the claims here. As before, I would lean toward questions rather than statements. I would include in those statements something about the AGF1 cluster, which doesn't really seem to me to be anything except "top right".
|
|
||
| # plot heatmap: | ||
| # non-spatial clusters are rows, and BANKSY with AGF clusters are columns | ||
| # non-spatial clusters are columns, and BANKSY with AGF clusters are rows |
There was a problem hiding this comment.
It would be very helpful to add axis labels with this information. Unfortunately, this is hard with pheatmap. As a workaround, you might add to the cluster labels so they are not just integers... e.g.. AGF1 AGF2, etc.
| patchwork::wrap_plots(plot_list, nrow = 2) | ||
| ``` | ||
|
|
||
| Do these patterns match your expectations from the coordinate plots showing clusters on the H&E? |
There was a problem hiding this comment.
I feel like we could use a bit more here? Personally, I'm not sure I have any expectations about a single gene in the context of clusters, and both seem to have gradients of expression...
Co-authored-by: Joshua Shapiro <josh.shapiro@ccdatalab.org>
Look, they don't even know what to do: https://prabhakarlab.github.io/Banksy/ I'm a bit tempted to just always call it |
Oh my. |
|
I went ahead and swapped all around to Copying the updated heatmap in here for review vs the whole notebook |
jashapiro
left a comment
jashapiro
left a comment
There was a problem hiding this comment.
Okay almost there! all the various edits mean that I get to revisit where we refer to the package vs. functions, which is minor, but there was one major thing:
The text still says we are not filtering to hvgs in this notebook, but we are! Sorry that I missed that in the last edit!
|
|
||
| `Banksy` first creates single matrix that is a weighted combination of the normalized expression in `logcounts` and the `Banksy` neighborhood matrix (`H0` or `H0` and `H1` if AGF is on). | ||
| (Note that `Banksy` will consider all genes present in the SPE when calculating the PCA; it does not subset to only the most variable genes. | ||
| We could perform this ourselves by subsetting the SPE before running `Banksy`, but for this notebook we will proceed with the full object.) |
There was a problem hiding this comment.
Wait, I thought we were filtering?
Co-authored-by: Joshua Shapiro <josh.shapiro@ccdatalab.org>
Co-authored-by: Joshua Shapiro <josh.shapiro@ccdatalab.org>
jashapiro
left a comment
jashapiro
left a comment
There was a problem hiding this comment.
LGTM (though one period was lost in the edit)
Co-authored-by: Joshua Shapiro <josh.shapiro@ccdatalab.org>
2 participants
Closes #993
This PR adds in remaining text to the BANKSY notebook.
I also modified a plot to be able to more easily see spots when interpreting expression.
However, I also am leaning towards removing the heatmap since I think it's just annoying to interpret and the expression plots do just as well. Wanted to get a second opinion though before I ripped it out though - let me know what you think!
02-spatial_clustering.nb.html