Damond_2019_Pancreas: Obtain the Damond_2019_Pancreas dataset

In BodenmillerGroup/imcdatasets: Collection of publicly available imaging mass cytometry (IMC) datasets

Obtain the Damond_2019_Pancreas dataset

Description

Obtain the Damond_2019_Pancreas dataset, which consists of three data objects: single cell data, multichannel images and cell segmentation masks. The data was obtained by imaging mass cytometry (IMC) of human pancreas sections from donors with type 1 diabetes.

Usage

Damond_2019_Pancreas(
  data_type = c("sce", "spe", "images", "masks"),
  full_dataset = FALSE,
  version = "latest",
  metadata = FALSE,
  on_disk = FALSE,
  h5FilesPath = NULL,
  force = FALSE
)

Arguments

data_type	type of object to load, 'images' for multichannel images or 'masks' for cell segmentation masks. Single cell data are retrieved using either 'sce' for the SingleCellExperiment format or 'spe' for the SpatialExperiment format.
full_dataset	if FALSE (default), a subset corresponding to 100 images is returned. If TRUE, the full dataset (corresponding to 845 images) is returned. Due to memory space limitations, this option is only available for single cell data and masks, not for data_type = "images".
version	dataset version. By default, the latest version is returned.
metadata	if FALSE (default), the data object selected in data_type is returned. If TRUE, only the metadata associated to this object is returned.
on_disk	logical indicating if images in form of HDF5Array objects (as .h5 files) should be stored on disk rather than in memory. This setting is valid when downloading images and masks.
h5FilesPath	path to where the .h5 files for on disk representation are stored. This path needs to be defined when on_disk = TRUE. When files should only temporarily be stored on disk, please set h5FilesPath = getHDF5DumpDir().
force	logical indicating if images should be overwritten when files with the same name already exist on disk.

Details

This is an Imaging Mass Cytometry (IMC) dataset from Damond et al. (2019):

• images contains a hundred 38-channel images in the form of a CytoImageList class object.

• masks contains the cell segmentation masks associated with the images, in the form of a CytoImageList class object.

• sce contains the single cell data extracted from the multichannel images using the cell segmentation masks, as well as the associated metadata, in the form of a SingleCellExperiment. This represents a total of 252,059 cells x 38 channels.

• spe same single cell data as for sce, but in the SpatialExperiment format.

All data are downloaded from ExperimentHub and cached for local re-use.

Mapping between the three data objects is performed via variables located in their metadata columns: mcols() for the CytoImageList objects and ColData() for the SingleCellExperiment and SpatialExperiment objects. Mapping at the image level can be performed with the image_name or image_number variables. Mapping between cell segmentation masks and single cell data is performed with the cell_number variable, the values of which correspond to the intensity values of the masks object. For practical examples, please refer to the "Accessing IMC datasets" vignette.

This dataset is a subset of the complete Damond et al. (2019) dataset comprising the data from three pancreas donors at different stages of type 1 diabetes (T1D). The three donors present clearly diverging characteristics in terms of cell type composition and cell-cell interactions, which makes this dataset ideal for benchmarking spatial and neighborhood analysis algorithms. If full_dataset = TRUE, the full dataset (845 images from 12 patients) is returned. This option is not available for multichannel images.

The assay slots of the SingleCellExperiment and SpatialExperiment objects contain three assays:

• counts contains raw mean ion counts per cell.

• exprs contains arsinh-transformed counts, with cofactor 1.

• quant_norm contains counts censored at the 99th percentile and scaled 0-1.

The marker-associated metadata, including antibody information and metal tags are stored in the rowData of the SingleCellExperiment / SpatialExperiment objects.

The cell-associated metadata are stored in the colData of the SingleCellExperiment and SpatialExperiment objects. These metadata include cell types (in colData(sce)$cell_type) and broader cell categories, such as "immune" or "islet" cells (in colData(sce)$cell_category). In addition, for cells located inside pancreatic islets, the islet they belong to is indicated in colData(sce)$islet_parent. For cells not located in islets, the "islet_parent" value is set to 0 but the spatially closest islet can be identified with colData(sce)$islet_closest.

The donor-associated metadata are also stored in the colData of the SingleCellExperiment and SpatialExperiment objects. For instance, the donors' IDs can be retrieved with colData(sce)$patient_id and the donors' disease stage can be obtained with colData(sce)$patient_stage.

Neighborhood information, defined here as cells that are localized next to each other, is stored as a SelfHits object in the colPairs slot of the SingleCellExperiment and SpatialExperiment objects.

The three donors in the subset present the following characteristics:

• 6126 is a non-diabetic donor, with large islets containing many beta cells, severe infiltration of the exocrine pancreas with myeloid cells but limited infiltration of islets.

• 6414 is a donor with recent T1D onset (shortly after diagnosis) showing partial beta cell destruction and mild infiltration of islets with T cells.

• 6180 is a donor with long-duration T1D (11 years after diagnosis), showing near-total beta cell destruction and limited immune cell infiltration in both the islets and the pancreas.

For information about other donors in the full dataset, please refer to the Damond et al. publication.

Dataset versions: a version argument can be passed to the function to specify which dataset version should be retrieved.

• `v0`: original version (Bioconductor <= 3.15).

• `v1`: consistent object formatting across datasets.

File sizes:

• `images`: size in memory = 7.4 Gb, size on disk = 1.7 Gb.

• `masks`: size in memory = 200 Mb, size on disk = 8.2 Mb.

• `sce`: size in memory = 353 Mb, size on disk = 204 Mb.

• `spe`: size in memory = 372 Mb, size on disk = 205 Mb.

• `sce_full`: size in memory = 2.4 Gb, size on disk = 1.5 Gb.

• `spe_full`: size in memory = 2.5 Gb, size on disk = 1.5 Gb.

• `masks_full`: size in memory = 1.4 Gb, size on disk = 60 Mb.

When storing images on disk, these need to be first fully read into memory before writing them to disk. This means the process of downloading the data is slower than directly keeping them in memory. However, downstream analysis will lose its memory overhead when storing images on disk.

Original source: Damond et al. (2019): https://doi.org/10.1016/j.cmet.2018.11.014

Original link to raw data, also containing the entire dataset: https://data.mendeley.com/datasets/cydmwsfztj/2

Value

A SingleCellExperiment object with single cell data, a SpatialExperiment object with single cell data, a CytoImageList object containing multichannel images, or a CytoImageList object containing cell segmentation masks.

Author(s)

Nicolas Damond

References

Damond N et al. (2019). A Map of Human Type 1 Diabetes Progression by Imaging Mass Cytometry. Cell Metab 29(3), 755-768.

Examples

# Load single cell data
sce <- Damond_2019_Pancreas(data_type = "sce")
print(sce)

# Display metadata
Damond_2019_Pancreas(data_type = "sce", metadata = TRUE)

# Load masks on disk
library(HDF5Array)
masks <- Damond_2019_Pancreas(data_type = "masks", on_disk = TRUE,
h5FilesPath = getHDF5DumpDir())
print(head(masks))

BodenmillerGroup/imcdatasets documentation built on March 20, 2024, 9:24 a.m.