Cluster: Clustering of samples
In debsin/dropClust: Single Cell Transcriptome Analysis

Description Usage Arguments Details Value Examples

Performs clustering on sampled cells and Post-hoc Cluster Assignment.

1
2
3

Cluster(object, use.subsamples = TRUE, method = "louvian",
  use.reduced.dims = NULL, k_nn = 10, conf = 0.75,
  use.previous = FALSE, ...)

`object`	A SingleCellExperiment object containing normalized expression values in `"normcounts"`.
`use.subsamples`	uses the indices obtained from the sampling method.
`method`	character, one of c("default", "hclust", "kmeans")., In the default mode, louvain based partition is used. When `hclust`, hierarchical clustering is used.
`use.reduced.dims`	optional, when used, the name of the `reducedDim()` table to be used for clustering. `method = "hclust"`.
`k_nn`	integers, specifies number of nearest neighbours, defaults to 10.
`conf`	numeric [0-1], defines the expected confidence of majority for a consensus. Cells remain unassigned when majority is below `conf`.
`use.previous`	optional, when `TRUE`, the clustering step is skipped and only the post-hoc clusering is repeated with the new `conf`.
`...`	For the `kmeans` method option, the argument `center` must be passed which specifies the number of clusters. Arguments specific to `cutreeDynamic` function may be passed when `method = "hclust"`; for reference, the default values of `minClusterSize` and `deepSplit` are set to 20 and 1 respectively.

Clustering is carried out in two alternate approaches on the sampled cells. For the default setting or quick identification of the existing broad clusters, a Louvain based partition is employed. Otherwise for fine-tuned clustering with outliers, hierarchical clustering is used with cutreeDynamic for dendrogram cut. Also, Assigns cluster membership to unsampled cells by using cluster membership information of the nearest neighbours. An approximate nearest neighbour graph is constructed out of the samples population using the find_ann() module. Some cells are left un-assigned when its neighbour's cluster membership doesn't form a majority as specified by the conf parameter. Unassigned cells (NA) are excluded in the plot or further downstream analysis.

List of:

cluster.ident vector cluster identifiers ranging from 1 to the number of clusters for respective data points.
nn.ids matrix, each row corresponds to a cell, whose columns depict cluster membership of its neighbours; as returned by the find_ann() module.

Unassigned samples are represented byNA values.

library(SingleCellExperiment)
ncells <- 1000
ngenes <- 2000
x <- matrix(rpois(ncells*ngenes, lambda = 10), ncol=ncells, nrow=ngenes, byrow=TRUE)
rownames(x) <- paste0("Gene", seq_len(ngenes))
colnames(x) <- paste0("Cell", seq_len(ncells))
sce <- SingleCellExperiment(list(counts=x))
sce <- CountNormalize(sce)
sce <- RankGenes(sce)
sce <- Cluster(sce,  use.subsamples=FALSE, conf=0.1)