AlignSubspace: Align subspaces using dynamic time warping (DTW)

Description

Aligns subspaces across a given grouping variable.

Usage

AlignSubspace(object, reduction.type = "cca", grouping.var, dims.align,
  num.possible.genes = 2000, num.genes = 30, show.plots = FALSE,
  verbose = TRUE, ...)

Arguments

object

Seurat object

reduction.type

Reduction to align scores for. Default is "cca".

grouping.var

Name of the grouping variable for which to align the scores

dims.align

Dims to align, default is all

num.possible.genes

Number of possible genes to search when choosing genes for the metagene. Set to 2000 by default. Lowering will decrease runtime but may result in metagenes constructed on fewer than num.genes genes.

num.genes

Number of genes to use in construction of "metagene" (default is 30).

show.plots

Show debugging plots

verbose

Displays progress and other output

...

Additional parameters to ScaleData

Value

Returns Seurat object with the dims aligned, stored in object@dr$reduction.type.aligned

Details

Following is a description for the two group case but this can be extended to arbitrarily many groups which works by performing pairwise alignment to a reference group (the largest group). First, we identify genes that are driving variation in both datasets by looking at the correlation of gene expression with each projection vector (e.g. CC1) in both datasets. For this we use the biweight midcorrelation (bicor) and choose the top num.genes with the strongest bicor to construct a 'metagene' for each dataset. We then scale each metagene to match its 95% reference range and linearly shift them by the minimum difference between the two metagenes over the 10-90 quantile range. We then map each cell in the smaller dataset to a cell in the larger dataset using dynamic time warping (DTW) and apply the same map to the projection vectors ( CC vectors) to place both datasets on a common aligned scale. We apply this procedue to each pair (group) of vectors individually for all specified in dims.align. For a full description of the method, see Butler et al 2017.

Examples

Run this code

# NOT RUN {
pbmc_small
# Requires CCA to have previously been run
# As CCA requires two datasets, we will split our test object into two just for this example
pbmc1 <- SubsetData(pbmc_small,cells.use = pbmc_small@cell.names[1:40])
pbmc2 <- SubsetData(pbmc_small,cells.use = pbmc_small@cell.names[41:80])
pbmc1@meta.data$group <- "group1"
pbmc2@meta.data$group <- "group2"
pbmc_cca <- RunCCA(pbmc1,pbmc2)
pbmc_cca <- AlignSubspace(pbmc_cca,reduction.type = "cca", grouping.var = "group", dims.align = 1:2)
# }
# NOT RUN {
# }

Run the code above in your browser using DataCamp Workspace