mrtree: Muti-resolution reconciled tree (MRtree)

In pengminshi/MRtree: Multi-resolution Reconciled Tree

Muti-resolution reconciled tree (MRtree)

Description

The method take as input the flat clusterings obtained across multiple resolution parameters. It reconciles the clusterings to produce a hierarchical cluster tree structure.

mrtree with label saved in a matrix as input

mrtree with labelmatrix save as a data frame as input

mrtree with SingleCellExperiment object as input

mrtree with Seurat object as input

Usage

mrtree(x, ...)

## S3 method for class 'matrix'
mrtree(
  x,
  prefix = NULL,
  suffix = NULL,
  max.k = Inf,
  consensus = FALSE,
  sample.weighted = FALSE,
  augment.path = FALSE,
  verbose = FALSE,
  n.cores = 1
)

## S3 method for class 'data.frame'
mrtree(x, prefix = NULL, suffix = NULL, ...)

## S3 method for class 'SingleCellExperiment'
mrtree(x, prefix = "sc3_", suffix = "_clusters", ...)

## S3 method for class 'Seurat'
mrtree(x, prefix = "RNA_snn_res.", suffix = NULL, ...)

Arguments

x	Seurat object
...	other parameters
prefix	srting indicating columns containing clustering information
suffix	string at the end of column names containing clustering information
max.k	the maximum resolution (number of clusters) to consider in building the tree
consensus	boolean, whether to perform consensus clustering within the clusterings with the same number of clusters.
sample.weighted	boolean, whether to weight the samples by the size of clusters, where higher weight is given to rare clusters. False by default.
augment.path	boolean, whether to augment one NA cluster in each layer to avoid shrinking the width of the tree with clustering alternate across layers. This will dramatically increase the running time. False by default.
verbose	boolean, whether to show running messages. False by default.
n.cores	integer, number of cores for parallel computation. 1 core by default.

Value

A list containing

labelmat.mrtree

The Reconciled tree saved as a label matrix, with duplicated layers omited.

labelmat.recon

The full reconciled tree save in a label matrix

labelmat.flat

The initial flat clustering (cluster tree) as input for MRtree algorithm

resolutions

The corresponding clustering resolution of the initial cluster tree

paths

The unique path in the resulted reconciled hierarchical cluster tree

Examples

data("clust_example")
# matrix as input
out = mrtree(clust_example$clusterings)
# out$labelmat.mrtree
# data frame with given prefix and suffix as input
df = as.data.frame(clust_example$clusterings); colnames(df) = paste0('K_', 1:4)
df$other = 1:nrow(clust_example$clusterings)  # add an additional column
out = mrtree(df, prefix = 'K_', suffix=NULL)

cl = cbind(clust_example$clusterings, clust_example$clusterings)
# add some additional noise
for (i in 1:ncol(cl)){
    cl[sample(10),i] = sample(1:length(unique(cl[,i])), 10, replace = TRUE)
}
mrtree(cl, consensus=TRUE) # mrtree with within-resolution consensus clustering
mrtree(cl, augment.path=TRUE) # mrtree with augmentation of NA cluster
mrtree(cl, sample.weighted=TRUE) # weight the sample by inverse of cluster size

# More example with Seurat and SingelCellExperiment object as input, see vignettes.

pengminshi/MRtree documentation built on March 6, 2023, 4:20 p.m.