Create SVG instruction notebook: code by jashapiro · Pull Request #1056 · AlexsLemonade/training-modules

jashapiro · 2026-08-03T20:32:26Z

Here I am adding the initial notebook for spatially variable genes as part of #1046

It is a bit rougher than I might have liked, but because I will be out for a few days, I wanted to get what I have up for initial comments as at least a draft.

As discussed, I am doing Moran's I & nnsvg analysis, with a little bit of intro to Voyager and SpatialFeatureExperiment structure to start out. I also have some comparisons between the two methods, and looking at SVGs vs. biological variation, which I want to talk a bit more about (how might technical variation be spatial?)

I have not yet decided which chunks will be live and not, so if you have particular ones that you think absolutely should or should not be, please feel free to comment on that!

Here is a rendered version for easier viewing.
04-spatially_variable_genes.nb.html

…instruction

sjspielman

Overall seems like a solid draft, and I left comments for recommended live chunks.

The main concern I have is doubling down on comparisons for p-values that we know aren't going to be very good, with mcmc. This seems especially tricky if we have concerns that the moran's test pvalues aren't robust. We can tackle both at once with some pre-computed pvalues from a thorough mcmc and both see: here's how you would actually get solid mcmc pvalues, and from these we can maybe get a sense of how robust the moran's test once are.

Otherwise, I'm not convinced that the comparisons are meaningful to interpret at all since they rely on us trusting the pvalues.

sjspielman · 2026-08-05T14:38:26Z

+You'll notice that the new column names include the sample_id, which is a bit annoying, but it is necessary to keep things separate if we have multiple samples.
+
+Let's first look at the distribution of Moran's I values across all (highly variable) genes:
+


Let's toss in a chunk to update names() here for typo reduction later?

sjspielman · 2026-08-05T14:44:49Z

+
+```{r moran_significant}
+moran_df <- moran_df |>
+  dplyr::mutate(
+    moran_test_p_adjust = p.adjust(moran_test_p, method = "BH"),


I might make an implementation comment here that we need to do this on an ungrouped data frame for the correction to work properly

and other minor changes

…instruction

jashapiro · 2026-08-11T21:05:55Z

Thanks for the review!

I have now added a small script and updates to the Snakemake files to pre-calculate the Moran Monte Carlo statistics with 100K replicates, which we now load in separately for a bit more comparison. Doing it this way allow seeing that the parametric statistic overestimates significance compared to the MC method (though probably not too bad, all things considered).

I did not yet update to include the actual output file in the analysis directory, mostly because I could not decide whether to put it in the repo or only on S3. The file is not very large (215KB), and could be much smaller if we stored it gzipped (~30KB). So part of me wants to just keep it in the repo, but that also seems kind of silly.

Other changes include small updates to plots, one of which is adding the correlation stats at the end directly to the plots with the annotate() function. Just for fun.

Attached is an updated rendered notebook:
04-spatially_variable_genes.nb.html

sjspielman · 2026-08-18T14:40:55Z

@jashapiro it doesn't look like the HTML has output from the chunks.. are you able to regenerate it? Thanks!

sjspielman

Overall I think this looks great, no major comments. I would like to see a rendered notebook though before I approve to look that over in case there's something not apparent here.

sjspielman · 2026-08-18T14:44:34Z

+  make_option(
+    opt = c("-t", "--threads"),
+    type = "integer",
+    default = 8,


I might make this default 4, since we're overriding it in practice anyways. This doesnt matter too much though.

Suggested change

default = 8,

default = 4,

sjspielman · 2026-08-18T14:49:58Z

FYI I opened #1064

sjspielman · 2026-08-18T15:09:17Z

+  dplyr::count(moran_test_sig, moran_mc_sig)
+```
+
+Either way, we get the vast majority of genes as "significant" which is not too surprising.


Why isn't this surprising, just because we might expect that HVGs overlap with SVGs?

Mostly because we can see from previous analysis that there is a lot of spatial variation in expression patterns. Once we condition on having high levels of variation, it seems almost certain that it will be spatially variable.

sjspielman · 2026-08-18T15:10:49Z

+Again, most of them!
+So most genes are significantly spatially variable, which is perhaps not a huge surprise.
+
+Other stats we can look at include `prop_sv`, which is the proportion of variation that is attributable to spatial variation.


breadcrumb for future text - does nnSVG model other kinds of variation, or is just spatial + residual? I don't remember myself...

sjspielman · 2026-08-18T15:12:01Z

+    label = paste0(
+      "Spearman correlation: ", 
+      cor(svg_df$moran_i, svg_df$nnsvg_propsv, method = 'spearman')
+    )


this is a great case where I agree with paste and not glue! sometimes it happens 😉

jashapiro · 2026-08-18T16:51:07Z

it doesn't look like the HTML has output from the chunks.. are you able to regenerate it? Thanks!

Sorry about that. This one should:
04-spatially_variable_genes.nb.html

Co-authored-by: Stephanie J. Spielman <stephanie.spielman@gmail.com> Co-authored-by: Joshua Shapiro <josh.shapiro@ccdatalab.org>

sjspielman

Rendered notebook looks good! Just caught a few small plot styling spots.

Co-authored-by: Stephanie J. Spielman <stephanie.spielman@gmail.com>

jashapiro · 2026-08-19T17:34:52Z

Thanks for the review.

I added TODOs for text areas I want to be sure to address.

One question I asked above and did want to resolve in this PR was the following, if you have thoughts:

I did not yet update to include the actual output file in the analysis directory, mostly because I could not decide whether to put it in the repo or only on S3. The file is not very large (215KB), and could be much smaller if we stored it gzipped (~30KB). So part of me wants to just keep it in the repo, but that also seems kind of silly.

sjspielman

I did not yet update to include the actual output file in the analysis directory, mostly because I could not decide whether to put it in the repo or only on S3. The file is not very large (215KB), and could be much smaller if we stored it gzipped (~30KB). So part of me wants to just keep it in the repo, but that also seems kind of silly.

Woops, thanks for the ping on this comment! I think it's fine to just keep it in S3.

Everything looks good to me, except one big comment that we are missing a seed in the MCMC script which seems important to have and regenerate the tsv with! I don't need to see again though.

sjspielman · 2026-08-19T18:41:57Z

+This includes all of the spots where we have data.
+
+
+## Identify Spatially Variable Genes


parallel (but feel free to ignore this since headers may be changing with the next PR anyways!)

Suggested change

## Identify Spatially Variable Genes

## Identifying Spatially Variable Genes

sjspielman · 2026-08-19T18:43:23Z

Just realized this script is missing a seed! We definitely want that...

…into jashapiro/1046-svg-instruction

jashapiro

realized I left a bunch of comments pending

jashapiro · 2026-08-19T17:17:07Z

+### Moran's I
+
+The measure of spatial covariation we will use is Moran's I.


I'll make it italics, which doesn't really help, I know but I don't want to make it code, since it isn't that.

jashapiro · 2026-08-19T17:17:19Z

+sfe <- Voyager::runMoransI(sfe)
+```
+
+This gives us an I value for each gene as well as the kurtosis of the gene's expression (how fat the tails are) as a value `K`.


added a todo

jashapiro · 2026-08-19T17:21:01Z

+We could do better with more replicates, but that takes a very long time to run... so we pre-calculated a set with 100,000 replicates.
+
+```{r read-moran_mc100k}
+moran_mc100k_df <- readr::read_tsv(moran_mc100k_tsv) |>


I'd rather read it here. While it doesn't matter in this case, there are often good (memory) reasons not to load all data at the top. While I think we tend to load things at the top most of the time, I don't think it is something we specifically call out for structure.

jashapiro · 2026-08-19T17:25:45Z

+  dplyr::count(moran_test_sig, moran_mc_sig)
+```
+
+Either way, we get the vast majority of genes as "significant" which is not too surprising.


Mostly because we can see from previous analysis that there is a lot of spatial variation in expression patterns. Once we condition on having high levels of variation, it seems almost certain that it will be spatially variable.

jashapiro · 2026-08-19T18:55:23Z

jashapiro added 3 commits August 3, 2026 16:22

Add initial svg notebook

90b2497

Merge remote-tracking branch 'origin/master' into jashapiro/1046-svg-…

1f8cacf

…instruction

fix title

6fb23e8

jashapiro assigned sjspielman Aug 3, 2026

jashapiro added 2 commits August 3, 2026 16:44

spelling updates

31f0aa3

missed a couple

0f8b2a4

sjspielman assigned sjspielman and unassigned sjspielman Aug 4, 2026

sjspielman self-requested a review August 4, 2026 12:20

sjspielman removed their assignment Aug 4, 2026

sjspielman reviewed Aug 5, 2026

View reviewed changes

jashapiro added 8 commits August 10, 2026 10:56

set live chunks

e152142

and other minor changes

Merge remote-tracking branch 'origin/master' into jashapiro/1046-svg-…

808a673

…instruction

Add calculation of moran mc results

0465f4a

add pre-computed replicates

0073419

Use labels for correlations

9ba9508

rename snakefile variables

8d5dc15

ignore unneeded snakemake output

df8b02b

typo fix

8688b27

jashapiro commented Aug 11, 2026

View reviewed changes

Comment thread spatial/setup/osteo/Snakefile Outdated

fix snakemake path

acbaf0c

jashapiro requested a review from sjspielman August 11, 2026 21:05

jashapiro marked this pull request as ready for review August 14, 2026 20:23

sjspielman mentioned this pull request Aug 18, 2026

Fill in the spatial cooking show magic file #1064

Open

sjspielman reviewed Aug 18, 2026

View reviewed changes

jashapiro commented Aug 18, 2026

View reviewed changes

Comment thread spatial/setup/osteo/calculate-moran.R

jashapiro commented Aug 18, 2026

View reviewed changes

Comment thread spatial/04-spatially_variable_genes.Rmd Outdated

Apply suggestions from code review

64c39f3

Co-authored-by: Stephanie J. Spielman <stephanie.spielman@gmail.com> Co-authored-by: Joshua Shapiro <josh.shapiro@ccdatalab.org>

sjspielman reviewed Aug 18, 2026

View reviewed changes

Comment thread spatial/04-spatially_variable_genes.Rmd Outdated

Comment thread spatial/04-spatially_variable_genes.Rmd Outdated

Comment thread spatial/04-spatially_variable_genes.Rmd Outdated

jashapiro and others added 5 commits August 18, 2026 15:11

Apply suggestions from code review

4ceba79

Co-authored-by: Stephanie J. Spielman <stephanie.spielman@gmail.com>

Merge branch 'master' into jashapiro/1046-svg-instruction

d1c9f18

add todos

a72a9fe

update qs2 ref name

d0fd09a

Merge branch 'master' into jashapiro/1046-svg-instruction

85f3108

jashapiro requested a review from sjspielman August 19, 2026 17:34

sjspielman approved these changes Aug 19, 2026

View reviewed changes

jashapiro added 2 commits August 19, 2026 14:53

add seed

5fb3673

Merge remote-tracking branch 'origin/jashapiro/1046-svg-instruction' …

71b2bf8

…into jashapiro/1046-svg-instruction

jashapiro commented Aug 19, 2026

View reviewed changes

jashapiro merged commit 77c81d0 into master Aug 20, 2026
5 checks passed

jashapiro deleted the jashapiro/1046-svg-instruction branch August 20, 2026 12:03

		You'll notice that the new column names include the sample_id, which is a bit annoying, but it is necessary to keep things separate if we have multiple samples.

		Let's first look at the distribution of Moran's I values across all (highly variable) genes:

		This includes all of the spots where we have data.


		## Identify Spatially Variable Genes

	## Identify Spatially Variable Genes
	## Identifying Spatially Variable Genes

		### Moran's I

		The measure of spatial covariation we will use is Moran's I.

Uh oh!

Uh oh!

Conversation

jashapiro commented Aug 3, 2026

Uh oh!

sjspielman left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

jashapiro commented Aug 11, 2026

Uh oh!

sjspielman commented Aug 18, 2026

Uh oh!

sjspielman left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jashapiro commented Aug 18, 2026

Uh oh!

Uh oh!

Uh oh!

sjspielman left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

jashapiro commented Aug 19, 2026

Uh oh!

sjspielman left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jashapiro left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!