GetCellMat: Extract real cell matrix from 'CB2FindCell' output and...

In zijianni/scCB2: CB2 improves power of cell detection in droplet-based single-cell RNA sequencing data

Extract real cell matrix from CB2FindCell output and optionally filter out low-quality cells

Description

Handy function to extract real cell matrix from CB2FindCell output. It provides the option to filter out broken cells based on proportion of mitochondrial gene expressions. The input can also be a sparse matrix only for cell filtering.

Usage

GetCellMat(CBout, MTfilter = 0.25, MTgene = NULL)

Arguments

CBout	Output object from CB2FindCell, or a sparse matrix (for example, from QuickCB2).
MTfilter	Numeric value between 0 and 1. Default: 0.25. For each barcode, if the proportion of mitochondrial gene expression exceeds MTfilter, this barcode will be filtered out. By default, cell barcodes with more than 25 filtered out. Set MTfilter = 1 for no filtering. The proportion of mitochondrial gene expressions is usually a criterion for evaluating cell quality, and is calculated using the scaled sum of all genes starting with "MT-" (human) or "mt-" (mouse) if row names are gene symbols, or customized mitochondrial genes specified by MTgene.
MTgene	Character vector. User may specify customized mitochondrial gene names to perform the filtering. This should correspond to a subset of row names in raw data.

Value

A dgCMatrix count matrix of real cells.

Examples

# Please also refer to the example in function CB2FindCell.

# Simulate CB2FindCell output object.
library(SummarizedExperiment)
data(mbrainSub)
mbrainReal <- mbrainSub[,Matrix::colSums(mbrainSub)>500]

CBOut <- SummarizedExperiment(
    list(cell_matrix = mbrainReal[,sample(ncol(mbrainReal), 
                                          200, replace = TRUE)]))
                 
# Get cell matrix, filtering out barcodes with 
# more than 10% of counts from mitochondrial genes.     

RealCell <- GetCellMat(CBOut, MTfilter = 0.1)
str(RealCell)             

zijianni/scCB2 documentation built on April 24, 2023, 12:55 p.m.