hierarchicalSplit {ADAPTS}

Build hierarchical cell clusters.

Description

Attempt to deconvolve cell types by building a hierarchy of cell types using spillToConvergence to determine cell types that are not signficantly different. First deconvolve those clusters of cell types. Deconvolution matrices are then built to separate the cell types that formerly could not be separated.

Usage

hierarchicalSplit(
  sigMatrix,
  geneExpr,
  oneCore = FALSE,
  nPasses = 100,
  deconMatrices = NULL,
  remZinf = TRUE,
  method = "DCQ",
  useRF = TRUE,
  incNonCluster = TRUE
)

Arguments

sigMatrix

The deconvolution matrix, e.g. LM22 or MGSM27

geneExpr

The source gene expression matrix used to calculate sigMatrix

oneCore

Set to TRUE to disable parallelization (DEFAULT: FALSE)

nPasses

The maximum number of iterations for spillToConvergence (DEFAULT: 100)

deconMatrices

Optional pre-computed results from spillToConvergence (DEFAULT: NULL)

remZinf

Set to TRUE to remove any ratio with zero or infinity when generating gList (DEFAULT: FALSE)

method

One of 'DCQ', 'SVMDECON', 'DeconRNASeq', 'proportionsInAdmixture', 'nnls' (DEFAULT: DCQ)

useRF

Set to TRUE to use ranger random forests to build the seed matrix (DEFAULT: TRUE)

incNonCluster

Set to TRUE to include a 'nonCluster' in each of the sub matrices (DEFAULT: TRUE)

Value

A list of clusters and a list of signature matrices for breaking those clusters

Examples

#This toy example 
library(ADAPTS)
fullLM22 <- ADAPTS::LM22[1:30, 1:4]
smallLM22 <- fullLM22[1:25,] 

clusters <- hierarchicalSplit(sigMatrix=smallLM22, geneExpr=fullLM22, oneCore=TRUE, nPasses=10,
    deconMatrices=NULL, remZinf=TRUE, method='DCQ', useRF=TRUE, incNonCluster=TRUE)