Init spatial experiment

[keviny2]

No description provided.

[jkanche]

Looks great! Awesome work!

[jkanche]

Just remembered, slicing needs to be updated so it also slices the image and spatial coords props. Technically only need to implement this function similar to SingleCellExperiment here

[keviny2]

Recording behavior regarding img_data and column_data.

Constructor

• All sample_ids in img_data must be present in column_data['sample_id'], but not all sample_ids in column_data need to be present in img_data['sample_id'] (i.e. there can be sample_ids in column_data['sample_id'] that are not in img_data['sample_id']

column_data setter

• Bioconductor only checks that the number of unique sample_ids in column_data remains constant, irrespective of the actual values of the new sample_ids
• It is possible to set a new column_data with a completely different set of sample_ids (doesn't even have to include the sample_id in img_data); as long as the number of unique sample_ids in the new column_data match the old column_data. Confusingly, doing this this will actually modify the sample_id in the img_data...

Here's an example

library(SpatialExperiment)
dir <- system.file(
   file.path("extdata", "10xVisium", "section1", "outs"),
   package = "SpatialExperiment")

# read in counts
fnm <- file.path(dir, "raw_feature_bc_matrix")
sce <- DropletUtils::read10xCounts(fnm)

# read in image data
img <- readImgData(
    path = file.path(dir, "spatial"),
    sample_id = "foo")

# read in spatial coordinates
fnm <- file.path(dir, "spatial", "tissue_positions_list.csv")
xyz <- read.csv(fnm, header = FALSE,
    col.names = c(
        "barcode", "in_tissue", "array_row", "array_col",
        "pxl_row_in_fullres", "pxl_col_in_fullres"))
xyz$sample_id <- c(rep("bar",25),rep("foo",25))

# construct observation & feature metadata
rd <- S4Vectors::DataFrame(
    symbol = rowData(sce)$Symbol)

# construct 'SpatialExperiment'
spe <- SpatialExperiment(
    assays = list(counts = assay(sce)),
    rowData = rd, 
    colData = DataFrame(xyz), 
    spatialCoordsNames = c("pxl_col_in_fullres", "pxl_row_in_fullres"),
    imgData = img
)

xyz_new <- DataFrame(xyz)
xyz_new$sample_id <- c(rep("bar",25),rep("baz",25))
colData(spe) <- xyz_new

# no more `foo` in `sample_id`
colData(spe)

# `sample_id` changes to `baz`
imgData(spe)

@jkanche How do you think we should handle this?

[jkanche]

I can understand why this might be the case (if the image is huge or someone doesn't care too much about the image itself). So I propose In addition to the current check you already have in the constructor, we

Rule (1): check if all sample_ids in column_data are accounted for in imgdata["sample_id"] and if they do not, we don't raise an exception but warn the user. We may in the future after talking to the SPE folks can decide to change this to an Exception rather than a warning.

Constructor

• All sample_ids in img_data must be present in column_data['sample_id'], but not all sample_ids in column_data need to be present in img_data['sample_id'] (i.e. there can be sample_ids in column_data['sample_id'] that are not in img_data['sample_id']

I don't understand the checking for unique number of sample_ids. So to be consistent, what makes sense to me right now is we do the same as what we do in the constructor: all sample_ids in img_data need to be in column_data's sample_id column and apply Rule (1).

column_data setter

* Bioconductor only checks that the number of unique `sample_id`s in `column_data` remains constant, irrespective of the actual values of the new `sample_id`s

* It is possible to set a new `column_data` with a completely different set of `sample_id`s (doesn't even have to include the `sample_id` in `img_data`); as long as the number of unique `sample_id`s in the new `column_data` match the old `column_data`. Confusingly, doing this this will actually modify the `sample_id` in the `img_data`...

Does that make sense?

edit 1: If you also agree, lets document this behavior in the constructor and the column data setter methods.

[keviny2]

@jkanche I think the reason why the sample_ids don't have to match in the column_data setter is so that you can 'rename' samples. See this test case: https://github.com/drighelli/SpatialExperiment/blob/devel/tests/testthat/test_SpatialExperiment-colData.R#L20

See #8