runDR {CATALYST}R Documentation

Perform dim. reduction on a daFrame

Description

Wrapper function to perform dimensionality reduction methods on a daFrame object using scater.

Usage

runDR(x, ...)

## S4 method for signature 'daFrame'
runDR(x, dr = c("TSNE", "PCA", "MDS", "UMAP",
  "DiffusionMap"), rows_to_use = 1000, cols_to_use = NULL,
  overwrite = FALSE, ...)

Arguments

x

a daFrame.

...

additional parameters passed to the dim. reduction method (see runTSNE, runPCA, runMDS, runUMAP, and runDiffusionMap).

dr

character string specifying the dimensionaly reduction method.

rows_to_use

numeric vector of row indices (cells) to use. If NULL, all cells will be used. If a single integer value N, (default 1000) a subset of N cells will be drawn from each sample.

cols_to_use

character vector in colnames(x) or numeric vector of column indices to use for computing reduced dimensions.

overwrite

logical. Whether to force overwriting any existing dimension reduction of type dr.

Value

a daFrame with an additional entry titled "dr" in the reducedDims slot of the input daFrame.

Author(s)

Helena L. Crowell helena.crowell@uzh.ch

Examples

data(PBMC_fs, PBMC_panel, PBMC_md)
daf <- daFrame(PBMC_fs, PBMC_panel, PBMC_md)
daf <- cluster(daf)

# PCA on all cells
daf <- runDR(daf, "PCA")

# UMAP on 1000 random cells
daf <- runDR(daf, "UMAP", rows_to_use = sample(nrow(daf), 1e3))

reducedDims(daf)
head(reducedDim(daf, "UMAP"))

# PCA on 200 cells per sample
set.seed(1)
daf <- runDR(daf, "PCA", rows_to_use = 200, overwrite = TRUE)

# re-using PCA for t-SNE will fail when using different cells
## Not run: 
daf <- runDR(daf, "TSNE", rows_to_use = 1:500, use_dimred = "PCA")
## End(Not run)

# use same seed to assure the same subset of cells is sampled
set.seed(1)
daf <- runDR(daf, "TSNE", rows_to_use = 200, use_dimred = "PCA")

# number of rows used for each DR:
vapply(reducedDims(daf), nrow, numeric(1))

# running on subset can be done 2-ways
daf2 <- runDR(daf, "MDS", 1:100)
daf3 <- runDR(daf[1:100, ], "MDS")

# option 1 keeps object-dimensions
identical(dim(daf2), dim(daf))

# option 2 keeps only specified rows
all.equal(dim(daf3), c(100, ncol(daf)))

# reduced dimension are identical
identical(reducedDim(daf2), reducedDim(daf2))

# run t-SNE on B-cell clusters only
data(merging_table)
daf <- mergeClusters(daf, "meta20", merging_table, "merging")
cells_use <- grep("B-cells", cluster_ids(daf, "merging"))
daf <- runDR(daf, "TSNE", cells_use, overwrite = TRUE)
plotDR(daf, "TSNE", "merging")
[Package CATALYST version 1.8.7 Index]