dim_reduce: Dimension Reduction

In kgellatl/SignalCell: Network Analysis for single cell RNA sequencing Data (sc-RNASeq)

Description

This function will do dimensionality reduction.

Usage

dim_reduce(
  input,
  genelist,
  method,
  assay = NULL,
  nComp = 15,
  nVar = 0.85,
  log = F,
  scale = T,
  reducedDim_key = NULL,
  seed = 100
)

Arguments

input	the input sce
genelist	the subset of genes to perform dimensionality reduction on
method	the algorithm choice for reduction before tSNE (either "ICA", "PCA", "iPCA", or FALSE if you want to reuse).
assay	The assay to operate on. Will default to get_def_assay(input)
nComp	the number of components to reduce too before tSNE, 5-20 recommended.
nVar	cutoff for percent of variance explained from PCs
log	Whether or not to log the input assay
scale	Whether or not to scale the input assay
reducedDim_key	The name of the LEM within ReducedDim slot of SCE
seed	For tSNE, the seed. Can set to NULL if desired.

Details

If the method is ICA, independent component analysis will be performed, and then tSNE will do the final dimension reduction. If PCA is selected, PCA will be performed before on the expression matrix transpose before tSNE. This PCA will use the cells positions on the principal components. If iPCA is selected, PCA will be be performed but without transposing the data. This will create "meta cells" instead of meta genes created in the typical PCA. Then tSNE will be performed on each cells contribution (loading) to the meta cell. We find that iPCA is much more robust and leads to cleaner clusters than traditional PCA.

Examples

ex_sc_example <- dim_reduce(input = ex_sc_example, genelist = gene_subset, method = "iPCA", nComp = 15, tSNE_perp = 30, iterations = 500, print_progress=TRUE)

kgellatl/SignalCell documentation built on Sept. 3, 2020, 8:45 a.m.