R/normalize.R

In stela2502/BioData: The package allows to annotate a numeric data.table with col and row data

#' @name normalize
#' @aliases normalize,BioData-method
#' @docType methods
#' @description  normalize the expression data (sample wise) using DEseq2
#' @param x The BioData object (NGS expression data, not single cells)
#' @param readCounts The number of reads from each bam file or another value you want to normalize the data to
#' @param name the new name of the object (deafule old name + normalized)
#' @param force replace old norm data (FALSE)
#' @return the normalized data set (original data stored in NGS$raw
#' @title normalize a BioData::R6 object
#' @export normalize

#if ( ! isGeneric('normalize') ){ 
methods::setGeneric('normalize', ## Name
	function ( object, ... , name=NULL) { 
		standardGeneric('normalize')
	}
)
#}else {
#	print ("Onload warn generic function 'normalize' already defined - no overloading here!")
#}

setMethod('normalize', signature = c ('BioData'),
		definition = function (  object, readCounts=NULL, force=FALSE ,name=NULL) {
			if ( ! object$snorm ){
				
				if ( is.null(object$raw) ){
					object$raw <- object$dat
				}else {
					object$dat <- object$raw ## regen if e.g. force is used ;-)
				}
				if ( is.null( readCounts ) ) {
					readCounts <- as.vector( DESeq2::estimateSizeFactorsForMatrix ( as.matrix(object$raw)) )
				}
				object$samples$SizeFactor <- readCounts
				logThis(object)
				object$dat = FastWilcoxTest::NormalizeSamples( object$dat, readCounts )	
				#object$dat =  data.frame(t(apply(object$dat,1, function(a) { a / readCounts } ) ))
				colnames(object$dat) = colnames(object$raw)
				rownames(object$dat) = rownames(object$raw)

				object$snorm=TRUE
				if(is.null(name)){
					name = paste( object$name ,'DESeq2_normalized' )
				}
				object$name = name
				
				if ( object$logged) {
					object$logged = FALSE
					logThis(object) ## regen log
				}
			}
			invisible(object)
		})

#' @name normalize_R_buggy
#' @aliases normalize_R_buggy,BioData-method
#' @docType methods
#' @description  
#' normalize the expression data by subsampling as described in PMID 24531970
#' @param x The SingleCells::BioData::R6 object
#' @param reads the required read depth
#' @param name the name of the new object
#' @param  force re-normalize this object (default FALSE)
#' @return the normalized data set (original data stored in slot 'raw'
#' @title normalize a SingleCells::BioData::R6 object
#' @export normalize_R_buggy
if ( ! isGeneric('normalize_R_buggy') ){ methods::setGeneric('normalize_R_buggy', ## Name
			function ( object, ... , name=NULL) { 
				standardGeneric('normalize_R_buggy')
			}
	)
}else {
	print ("Onload warn generic function 'normalize_R_buggy' already defined - no overloading here!")
}
setMethod('normalize_R_buggy', signature = c ('SingleCells'),
		definition = function (  object, reads=600, force=FALSE , name=NULL) {
			if ( is.null( object$usedObj$snorm) ) {
				object$usedObj$snorm = 0
			}
			reads <- round(reads)
			
			if ( is.null(object$usedObj$snorm) )
				object$usedObj$snorm = 0
			if ( is.null(force))
				force = FALSE
			
			if ( force | object$usedObj$snorm == 0 ) {
				if ( length( object$samples$nUMI ) == 0 ) {
					object$samples$nUMI <- apply( object$dat, 2, sum)
				}
				
				if(is.null(name)){
					name = paste( object$name ,'resample_normalized' )
				}
				reduceTo( object, copy=FALSE, what="col", 
					to=colnames(object$dat)[which(object$samples$nUMI >= reads)],
					name=name )
				
				if ( is.null(object$raw) ){
					object$raw <- object$dat
				}
				## resample the data
				n <- nrow(object$raw)
				object$dat[] <- 0
				for ( i in 1:ncol(object$raw) ) {
					d <- sample(rep ( 1:n, object$raw[,i]) , reads, replace=T)
					t <- table(d)
					object$dat[ as.numeric(names(t)),i] <- as.numeric(t)
				}
				rownames(object$dat) <- rownames(object$raw)
				colnames(object$dat) <- colnames(object$raw)
			}
			else {
				print ("Data was already normalized - skipped")
			}
			object$usedObj$snorm = 1
			invisible(object)
		}
)


#' @describeIn normalize a SingleCells::BioData::R6 object using subsampling
#' @docType methods
#' @description normalize the expression data by subsampling as described in PMID 24531970
#' @param x The SingleCells::BioData::R6 object
#' @param reads the required read depth
#' @param name the name of the new object
#' @param  force re-normalize this object (default FALSE)
#' @return the normalized data set (original data stored in slot 'raw'
#' @title normalize a SingleCells::BioData::R6 object
#' @export normalize
setMethod('normalize', signature = c ('SingleCells'),
		definition = function (  object, reads=600, force=FALSE , name=NULL) {
			
			reads <- round(reads)
			
			if ( is.null(object$usedObj$snorm) )
				object$usedObj$snorm = 0
			if ( is.null(force))
				force = FALSE
			if ( force | object$usedObj$snorm == 0 ) {
				if ( length( object$samples$nUMI ) == 0 ) {
					object$samples$nUMI <- FastWilcoxTest::ColNotZero( object$dat )
				}
				
				if(is.null(name)){
					name = paste( object$name ,'resample_normalized' )
				}
				reduceTo( object, copy=FALSE, what="col", 
						to=colnames(object$dat)[which(object$samples$nUMI >= reads)],
						name=name )
				
				if ( is.null(object$raw) ){
					object$raw <- object$dat
				}
				
				object$dat = FastWilcoxTest::NormalizeCells( object$raw, reads )
				rownames(object$dat) <- rownames(object$raw)
				colnames(object$dat) <- colnames(object$raw)
			}
			else {
				print ("Data was already normalized - skipped")
			}
			object$usedObj$snorm = 1
			invisible(object)
		}
)



#' @describeIn normalize normalize a MicoArray::BioData::R6 object using quantile normalization
#' @docType methods
#' @description  constructor that has to be implemented for a generic BioData
#' This generic version was meant for array data and I have not had the need nor time to implement this part.
#' @param x the MicoArray::BioData::R6 object
#' @param to a numeric vector to normalize the samples to. Has to have the same length as are columns in the data table 
#' @title normalize a MicoArray::BioData::R6 object
#' @export normalize
setMethod('normalize', signature = c ('MicroArray') ,
	definition = function ( object , to=NULL, name=NULL) {
		object$zscored=NULL
		df_rank <- apply(object$dat,2,rank,ties.method="min")
		df_sorted <- data.frame(apply(object$data(), 2, sort))
		df_mean <- apply(df_sorted, 1, mean)
		
		index_to_mean <- function(my_index, my_mean){
			return(my_mean[my_index])
		}
		object$samples$norm_to = 'quantile'
		
		df_final <- apply(df_rank, 2, index_to_mean, my_mean=df_mean)
		rownames(df_final) <- rownames(df)
		
		object$dat = Matrix::Matrix( as.matrix(df_final) )
		
		invisible(object)
})