Dune

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

Dune(clusMat, ...)

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

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

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

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.
cluster_columns	if clusMat is a SummarizedExperiment, then this defines the columns of colData that are outputs from a clustering algorithm.

Value

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

Details

The Dune algorithm merges pairs of clusters in order to improve the mean adjusted Rand Index with other clustering labels. It returns a list with four 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

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

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$merges
#>   clusteringLabel cluster1 cluster2
#> 1               B       11       12
#> 2               C       11       12
#> 3               D       11       13
#> 4               E       11       13
#> 5               E       11       14
#> 6               D       11       12
#> 7               E       11       12
merger2$merges
#>   clusteringLabel cluster1 cluster2
#> 1               B       11       12
#> 2               C       11       12
#> 3               D       11       13
#> 4               E       11       13
#> 5               E       11       14
#> 6               D       11       12
#> 7               E       11       12