Skip to contents

Miscellaneous notes

Nicholas J. Eagles

Lieber Institute for Brain Development

Leonardo Collado-Torres

Lieber Institute for Brain DevelopmentCenter for Computational Biology, Johns Hopkins UniversityDepartment of Biostatistics, Johns Hopkins Bloomberg School of Public Health

24 October 2024

Source: vignettes/misc.Rmd
misc.Rmd

This vignette has some extra companion notes to the Introduction to visiumStiched main vignette.

Load data

Let’s load the spatialLIBD package we’ll use in this vignette.

Now we can download the example visiumStitched_brain data that includes normalized logcounts. We’ll define the same example white matter marker genes.

## Grab SpatialExperiment with normalized counts
spe <- fetch_data(type = "visiumStitched_brain_spe")
#> 2024-10-24 20:17:16.494094 loading file /github/home/.cache/R/BiocFileCache/760d7e2c6_visiumStitched_brain_spe.rds%3Frlkey%3Dnq6a82u23xuu9hohr86oodwdi%26dl%3D1

## Check that spe does contain the "logcounts" assay
assayNames(spe)
#> [1] "counts"    "logcounts"

## Define white matter marker genes
wm_genes <- rownames(spe)[
    match(c("MBP", "GFAP", "PLP1", "AQP4"), rowData(spe)$gene_name)
]

Geometric transformations notes

As a SpatialExperiment, the stitched data you constructed with visiumStitched::build_SpatialExperiment() may need to be rotated or mirrored by group. This can be done using the SpatialExperiment::rotateObject() or SpatialExperiment::mirrorObject() functions. These functions are useful in case the image needs to be transformed to reach the preferred tissue orientation.

## Rotate image and gene-expression data by 180 degrees, plotting a combination
## of white-matter genes
vis_gene(
    rotateObject(spe, sample_id = "Br2719", degrees = 180),
    geneid = wm_genes,
    assayname = "counts",
    is_stitched = TRUE,
    spatial = FALSE
)

## Mirror image and gene-expression data across a vertical axis, plotting a
## combination of white-matter genes
vis_gene(
    mirrorObject(spe, sample_id = "Br2719", axis = "v"),
    geneid = wm_genes,
    assayname = "counts",
    is_stitched = TRUE,
    spatial = FALSE
)

You might want to re-make these plots with spatial = TRUE so you can see how the histology image gets rotated and/or mirrored. For file size purposes of this vignette, here we had to use spatial = FALSE.

A note on normalization

As noted in the main vignette, library-size variation across spots can bias the apparent spatial distribution of genes when raw counts are used. The effect is often dramatic enough that spatial trends cannot be easily seen across the stitched data until data is log-normalized. Instead of performing normalization here, we’ll fetch the object with normalized counts from spatialLIBD, then plot a few white matter genes as before:

## Plot combination of normalized counts for some white-matter genes
vis_gene(
    spe,
    geneid = wm_genes,
    assayname = "logcounts",
    is_stitched = TRUE,
    spatial = FALSE
)

Recall the unnormalized version of this plot, which is not nearly as clean:

## Plot raw counts, which are noisier
## Same plot we made before, but this time with no histology images
vis_gene(
    spe,
    geneid = wm_genes,
    assayname = "counts",
    is_stitched = TRUE,
    spatial = FALSE
)

The actual normalization code for this example data is available here.

Merging overlapping spots

In general, we recommend retaining all spots for downstream analysis, even if that means including multiple spots per array coordinate. We show that many software tools, such as BayesSpace and PRECAST, can smoothly handle data in this format. However, given that having multiple spots at the same array coordinates is atypical in Visium experiments, we caution that it’s possible some software may break or not perform as intended with stitched data. We provide the merge_overlapping() function to address this case.

In particular, merge_overlapping() sums raw counts across spots that overlap, ultimately producing a SpatialExperiment with one spot per array coordinate. colData() information, both discrete and continuous, is taken from spots where exclude_overlapping is FALSE. Note that the function can be quite memory-intensive and time-consuming.

spe_merged <- merge_overlapping(spe)