Dune: Dune

In HectorRDB/Dune: Improving replicability in single-cell RNA-Seq cell type discovery

Description

Compute the Metric between every pair of clustering labels after merging every possible pair of clusters. Find the one that improves the Metric merging the most, merge the pair. Repeat until there is no improvement.

Usage

Dune(clusMat, ...)

## S4 method for signature 'matrix'
Dune(
  clusMat,
  unclustered = NULL,
  verbose = FALSE,
  parallel = FALSE,
  BPPARAM = BiocParallel::bpparam(),
  metric = "NMI"
)

## S4 method for signature 'data.frame'
Dune(
  clusMat,
  unclustered = NULL,
  verbose = FALSE,
  parallel = FALSE,
  BPPARAM = BiocParallel::bpparam(),
  metric = "NMI"
)

## S4 method for signature 'SummarizedExperiment'
Dune(
  clusMat,
  cluster_columns,
  unclustered = NULL,
  verbose = FALSE,
  parallel = FALSE,
  BPPARAM = BiocParallel::bpparam(),
  metric = "NMI"
)

Arguments

clusMat	the matrix of samples by clustering labels.
...	parameters including:
unclustered	The value assigned to unclustered cells. Default to NULL
verbose	Whether or not the print cluster merging as it happens.
parallel	Logical, defaults to FALSE. Set to TRUE if you want to parallellize the fitting.
BPPARAM	object of class bpparamClass that specifies the back-end to be used for computations. See bpparam in BiocParallel package for details. Won't be used if parallel is FALSE.
metric	The metric that is tracked to decide which clusters to merge. For now, either ARI and NMI are accepted. Default to NMI. See details.
cluster_columns	if clusMat is a SummarizedExperiment, then this defines the columns of colData that are outputs from a clustering algorithm.

Details

The Dune algorithm merges pairs of clusters in order to improve the mean adjusted Rand Index or the mean normalized mutual information with other clustering labels. It returns a list with five components.: #'

• initialMat: The initial matrix of cluster labels

• currentMat: The final matrix of cluster labels

• merges: The step-by-step detail of the merges, recapitulating which clusters where merged in which cluster label

• impMetric: How much each merge improved the mean Metric between the cluster label that has been merged and the other cluster labels.

• metric: The metric that was used to find the merges.

Value

A list with four components: the initial matrix of clustering labels, the final matrix of clustering labels, the merge info matrix and the Metric improvement vector.

See Also

clusterConversion ARIImp

Examples

data("clusMat", package = "Dune")
merger <- Dune(clusMat = clusMat)
# clusters 11 to 14 from cluster label 5 and 3 are subset of cluster 2 from
# other cluster labels. Designing cluster 2 as unclustered therefore means we
# do fewer merges.
merger2 <- Dune(clusMat = clusMat, unclustered = 2)
merger$merges
merger2$merges

HectorRDB/Dune documentation built on Dec. 18, 2020, 8:08 p.m.