2024 December updates

[sjspielman]

This issue tracks items we notice need fixing from the December 2024 advanced scRNA-seq workshop.

[sjspielman]

In the integration notebook, we should probably name the Y-axis proportion rather than its default count

training-modules/scRNA-seq-advanced/02-dataset_integration.Rmd

Line 657 in a53758e

# Use ggplot2 to make a barplot the cell types across samples

[sjspielman]

I didn't catch the full sentence Josh said during intro slides, but I caught the bit that we need to add an arrow somewhere in the "single sample roadmap" diagram (I believe related to marker gene or gene-set analysis).

[jaclyn-taroni]

This comment should say that it is the path to the Cell Ranger matrix directory:

training-modules/scRNA-seq-advanced/01-read_filter_normalize_scRNA.Rmd

Line 79 in a53758e

# Path to the Cell Ranger matrix file

[jaclyn-taroni]

The "Will it integrate?" slide in the integration slides should say "healthy and tumor" instead of "healthy and normal"

[jaclyn-taroni]

The first plotReducedDim() call in the integration notebook (line 463) has pretty complicated syntax -- I am not sure it's completely necessary that it is live.

[sjspielman]

In integration, I wonder if we want to change up a little of the opening chunks where we set file names and read them in. Some of these thoughts are based on (unrelated to training) code reviews that @jashapiro had left elsewhere, if he wants to weigh in on this potential change too:

• Rather than dir() we might use list.files() instead to list out what's in the data directory

  training-modules/scRNA-seq-advanced/02-dataset_integration.Rmd

  Lines 88 to 90 in a53758e

  	```{r input dir, live = TRUE}
  	dir(input_dir)
  	```

• Currently we use file.path() to form all the input file paths, but I wonder if it might simplify code to instead just go ahead and use list.files(full.names = TRUE) in the first place (or show list.files() first without and then with this argument)

  training-modules/scRNA-seq-advanced/02-dataset_integration.Rmd

  Lines 171 to 175 in a53758e

  	```{r define sce_paths, live = TRUE}
  	# Now, convert these to file paths: <input_dir>/<sample_name>.rds
  	sce_paths <- file.path(input_dir,
  	glue::glue("{sample_names}.rds")
  	)

• We add list names to sce_list only after we read in the files. We might want to add those names beforehand

  training-modules/scRNA-seq-advanced/02-dataset_integration.Rmd

  Lines 203 to 206 in a53758e

  	```{r add list names, live = TRUE}
  	# Assign the sample names as the names for sce_list
  	names(sce_list) <- sample_names
  	```

[sjspielman]

Suggested integration changes:

• We should beef up some of the explanation for how the cell types in Patel et al were obtained in the first place, since this helps contextualize how we use them to assess integration results

  training-modules/scRNA-seq-advanced/02-dataset_integration.Rmd

  Lines 213 to 216 in a53758e

  	If you look closely at the printed SCE objects, you may notice that they all contain `colData` table columns `celltype_fine` and `celltype_broad`.
  	These columns (which we added to SCE objects during [pre-processing](https://github.com/AlexsLemonade/training-modules/tree/master/scRNA-seq-advanced/setup/rms)) contain putative _cell type annotations_ as assigned in [Patel _et al._ (2022)](https://doi.org/10.1016/j.devcel.2022.04.003).
  	We will end up leveraging these cell type annotations to explore how successful our integration is; after integration, we expect cell types from different samples to group together, rather than being separated by batches.

• This code would make more sense with length(), not head()

  training-modules/scRNA-seq-advanced/02-dataset_integration.Rmd

  Lines 304 to 316 in a53758e

  	```{r shared genes}
  	# Define vector of shared genes
  	shared_genes <- sce_list |>
  	# get rownames (genes) for each SCE in sce_list
  	purrr::map(rownames) |>
  	# reduce to the _intersection_ among lists
  	purrr::reduce(intersect)
  	```
  	```{r print shared genes, live = TRUE}
  	# Use head to look at the vector of shared genes:
  	head(shared_genes)
  	```

• We should probably use sce or similar, not x, in these spots to emphasize that it's nice that you can have informative "loop variables" with this new(-ish) syntax
  

  training-modules/scRNA-seq-advanced/02-dataset_integration.Rmd

  Lines 290 to 294 in a53758e

  	```{r compare rowdata, live = TRUE}
  	# Use `purrr::map()` to quickly extract rowData column names for all SCEs
  	purrr::map(sce_list,
  	\(x) colnames(rowData(x)))
  	```

  
  

  training-modules/scRNA-seq-advanced/02-dataset_integration.Rmd

  Lines 334 to 337 in a53758e

  	```{r compare coldata}
  	purrr::map(sce_list,
  	\(x) colnames(colData(x)) )
  	```

• This code would make more sense with table(), not unique()

  training-modules/scRNA-seq-advanced/02-dataset_integration.Rmd

  Lines 427 to 428 in a53758e

  	# What are the unique values in the `sample` column?
  	unique( colData(merged_sce)$sample )

[jaclyn-taroni]

The DE dimension reduction plots by cell type:

training-modules/scRNA-seq-advanced/03-differential_expression.Rmd

Lines 226 to 234 in a53758e

	```{r celltype UMAP}
	# UMAP of all samples labeled by cell type
	scater::plotReducedDim(integrated_sce,
	dimred = "fastmnn_UMAP",
	# color each point by cell type
	color_by = "celltype_broad",
	point_size= 0.5,
	point_alpha = 0.4)
	```

Could probably use a tweak to the legend along the lines of what we have in the integration notebook:

training-modules/scRNA-seq-advanced/02-dataset_integration.Rmd

Line 471 in a53758e

guides(color = guide_legend(override.aes = list(size = 3, alpha = 1))) + # Modify the legend key with larger, easier to see points

To make the cell type colors easier to see when projected, etc.

[sjspielman]

This is not in fact the "last thing" we do in this notebook, but rather the "next thing":

training-modules/scRNA-seq-advanced/03-differential_expression.Rmd

Lines 630 to 631 in a53758e

	The last thing that we will do is take a look at how many genes are significant.
	Here we will want to use the adjusted p-value, found in the `padj` column of the results, as this accounts for multiple test correction.

[jashapiro]

Add set.seed() to setup in pathway analysis notebook

training-modules/scRNA-seq-advanced/04-gene_set_enrichment_analysis.Rmd

Line 40 in a53758e

## Set up

[jashapiro]

We don't use marker genes anymore, but differential expression analysis results:

training-modules/scRNA-seq-advanced/04-gene_set_enrichment_analysis.Rmd

Line 56 in a53758e

# We'll use the marker genes as GSEA input