RawDataCleaning: Raw data cleaning

In seriph78/COTAN: COexpression Tables ANalysis

Raw data cleaning

Description

These methods are to be used to clean the raw data. That is drop any number of genes/cells that are too sparse or too present to allow proper calibration of the COTAN model.

We call genes that are expressed in all cells Fully-Expressed while cells that express all genes in the data are called Fully-Expressing. In case it has been made quite easy to exclude the flagged genes/cells in the user calculations.

Usage

## S4 method for signature 'COTAN'
flagNotFullyExpressedGenes(objCOTAN)

## S4 method for signature 'COTAN'
flagNotFullyExpressingCells(objCOTAN)

## S4 method for signature 'COTAN'
getFullyExpressedGenes(objCOTAN)

## S4 method for signature 'COTAN'
getFullyExpressingCells(objCOTAN)

## S4 method for signature 'COTAN'
findFullyExpressedGenes(objCOTAN, cellsThreshold = 0.99)

## S4 method for signature 'COTAN'
findFullyExpressingCells(objCOTAN, genesThreshold = 0.99)

## S4 method for signature 'COTAN'
dropGenesCells(
  objCOTAN,
  genes = vector(mode = "character"),
  cells = vector(mode = "character")
)

ECDPlot(objCOTAN, yCut)

## S4 method for signature 'COTAN'
clean(
  objCOTAN,
  cellsCutoff = 0.003,
  genesCutoff = 0.002,
  cellsThreshold = 0.99,
  genesThreshold = 0.99
)

cleanPlots(objCOTAN, includePCA = TRUE)

cellSizePlot(objCOTAN, splitPattern = " ", numCol = 2L)

genesSizePlot(objCOTAN, splitPattern = " ", numCol = 2L)

mitochondrialPercentagePlot(
  objCOTAN,
  splitPattern = " ",
  numCol = 2L,
  genePrefix = "^MT-"
)

scatterPlot(objCOTAN, splitPattern = " ", numCol = 2L, splitSamples = FALSE)

Arguments

objCOTAN	a COTAN object
cellsThreshold	any gene that is expressed in more cells than threshold times the total number of cells will be marked as fully-expressed. Default threshold is 0.99 \; (99.0\%)
genesThreshold	any cell that is expressing more genes than threshold times the total number of genes will be marked as fully-expressing. Default threshold is 0.99 \; (99.0\%)
genes	an array of gene names
cells	an array of cell names
yCut	y threshold of library size to drop
cellsCutoff	clean() will delete from the raw data any gene that is expressed in less cells than threshold times the total number of cells. Default cutoff is 0.003 \; (0.3\%)
genesCutoff	clean() will delete from the raw data any cell that is expressing less genes than threshold times the total number of genes. Default cutoff is 0.002 \; (0.2\%)
includePCA	a Boolean flag to determine whether to calculate the PCA associated with the normalized matrix. When TRUE the first four elements of the returned list will be NULL
splitPattern	Pattern used to extract, from the column names, the sample field (default " ")
numCol	Once the column names are split by splitPattern, the column number with the sample name (default 2)
genePrefix	Prefix for the mitochondrial genes (default "^MT-" for Human, mouse "^mt-")
splitSamples	Boolean. Whether to plot each sample in a different panel (default FALSE)

Details

flagNotFullyExpressedGenes() returns a Boolean array with TRUE for those genes that are not fully-expressed.

flagNotFullyExpressingCells()returns a Boolean vector with TRUE for those cells that are not expressing all genes

getFullyExpressedGenes() returns the genes expressed in all cells of the dataset

getFullyExpressingCells() returns the cells that did express all genes of the dataset

findFullyExpressedGenes() determines the fully-expressed genes inside the raw data

findFullyExpressingCells() determines the cells that are expressing all genes in the dataset

dropGenesCells() removes an array of genes and/or cells from the current COTAN object.

ECDPlot() plots the empirical distribution function of library sizes (UMI number). It helps to define where to drop "cells" that are simple background signal.

clean() is the main method that can be used to check and clean the dataset. It will discard any genes that has less than 3 non-zero counts per thousand cells and all cells expressing less than 2 per thousand genes. It also produces and stores the estimators for nu and lambda

cleanPlots() creates the plots associated to the output of the clean() method.

cellSizePlot() plots the raw library size for each cell and sample.

genesSizePlot() plots the raw gene number (reads > 0) for each cell and sample

mitochondrialPercentagePlot() plots the raw library size for each cell and sample.

scatterPlot() creates a plot that check the relation between the library size and the number of genes detected.

Value

flagNotFullyExpressedGenes() returns a Booleans array with TRUE for genes that are not fully-expressed

flagNotFullyExpressingCells() returns an array of Booleans with TRUE for cells that are not expressing all genes

getFullyExpressedGenes() returns an array containing all genes that are expressed in all cells

getFullyExpressingCells() returns an array containing all cells that express all genes

findFullyExpressedGenes() returns the given COTAN object with updated fully-expressed genes' information

findFullyExpressingCells() returns the given COTAN object with updated fully-expressing cells' information

dropGenesCells() returns a completely new COTAN object with the new raw data obtained after the indicated genes/cells were expunged. All remaining data is dropped too as no more relevant with the restricted matrix. Exceptions are:

• the meta-data for the data-set that gets kept unchanged

• the meta-data of genes/cells that gets restricted to the remaining elements. The columns calculated via estimate and find methods are dropped too

ECDPlot() returns an ECD plot

clean() returns the updated COTAN object

cleanPlots() returns a list of ggplot2 plots:

• "pcaCells" is for pca cells

• "pcaCellsData" is the data of the pca cells (can be plotted)

• "genes" is for B group cells' genes

• "UDE" is for cells' UDE against their pca

• "nu" is for cell nu

• "zoomedNu" is the same but zoomed on the left and with an estimate for the low nu threshold that defines problematic cells

cellSizePlot() returns the violin-boxplot plot

genesSizePlot() returns the violin-boxplot plot

mitochondrialPercentagePlot() returns a list with:

• "plot" a violin-boxplot object

• "sizes" a sizes data.frame

scatterPlot() returns the scatter plot

Examples

library(zeallot)

data("test.dataset")
objCOTAN <- COTAN(raw = test.dataset)

genes.to.rem <- getGenes(objCOTAN)[grep('^MT', getGenes(objCOTAN))]
cells.to.rem <- getCells(objCOTAN)[which(getCellsSize(objCOTAN) == 0)]
objCOTAN <- dropGenesCells(objCOTAN, genes.to.rem, cells.to.rem)

objCOTAN <- clean(objCOTAN)

objCOTAN <- findFullyExpressedGenes(objCOTAN)
goodPos <- flagNotFullyExpressedGenes(objCOTAN)

objCOTAN <- findFullyExpressingCells(objCOTAN)
goodPos <- flagNotFullyExpressingCells(objCOTAN)

feGenes <- getFullyExpressedGenes(objCOTAN)

feCells <- getFullyExpressingCells(objCOTAN)

## These plots might help to identify genes/cells that need to be dropped
ecdPlot <- ECDPlot(objCOTAN, yCut = 100)
plot(ecdPlot)

# This creates many infomative plots useful to determine whether
# there is still something to drop...
# Here we use the tuple-like assignment feature of the `zeallot` package
c(pcaCellsPlot, ., genesPlot, UDEPlot, ., zNuPlot) %<-% cleanPlots(objCOTAN)
plot(pcaCellsPlot)
plot(UDEPlot)
plot(zNuPlot)

lsPlot <- cellSizePlot(objCOTAN)
plot(lsPlot)

gsPlot <- genesSizePlot(objCOTAN)
plot(gsPlot)

mitPercPlot <-
  mitochondrialPercentagePlot(objCOTAN, genePrefix = "g-0000")[["plot"]]
plot(mitPercPlot)

scPlot <- scatterPlot(objCOTAN)
plot(scPlot)

seriph78/COTAN documentation built on May 2, 2024, 11:17 a.m.