Deconvolution: Deconvolution function

In chichaumiau/DSA: Digital sorting of complex tissues for cell type specific gene expression profiles.

Description

Function to estimate the deconvoluted signals - deconvolve the mixture signals to cell-type specific signals for each gene.

Usage

Deconvolution(data, weight, method = "LM", l = 0, u = 2^34)

Arguments

data	data matrix of the mixture signals, with genes in row, cell type in column, in anti-log scale.
weight	weight matrix, with cell type in row, tissue types in column.
method	methods used in estimating true signals for each tissue type. Default to 'LM' for linear regression. Other methods included 'LG' (logistic regression),'QP_LM' or 'QP_LG' (quadradic programming with constraint on the estimated parameter on linear/logistic regression) For methods 'LG' and 'QP_LG', input data is transformed into log-scaled and returned values are anti-logged.
l,u	values for the lower- (l) and upper- (u) bound used in setting the vector for values of b0 (bvec for solve.QP) in solving for quadratic programming. Defaults to 0 and 2^34 respectively.

Value

paraM

a matrix of deconvoluted signals (anti-log scale) with genes in row, tissue types in columns

See Also

EstimateWeight, GSM_QP, solve.QP,

Examples

	## load package DSA
	library(DSA)
	
	## load sample data
	data('mix.signals')
	data('cell.gene')

	# In this data, thr first three samples contain signal from only one cell	
	pure <- mix[, 1:3]
	mix <- mix[, 4:14]

	# Generate a list object set of marker genes for each cell type
	unique_cell_type <- unique(cell.gene[,2]) 
	gene_list <- list()
	for( i in 1 : length(unique_cell_type)){
		gene_list[[i]] <- cell.gene[cell.gene[,2] == unique_cell_type[i],1]
	} 
	names(gene_list) <- unique_cell_type

	# Estimate weight matrix
	estimated_weight <- EstimateWeight(2^mix, gene_list, method="LM")
	
	# Estimate deconvoluted signals
	deconv <- as.matrix(Deconvolution(2^mix, t(estimated_weight$weight), method="QP_LM", l=min(2^mix), u=max(2^mix)))
	rownames(deconv) <- rownames(mix)
	colnames(deconv) <- unique_cell_type
	
	# check the estimated cell-type specific signals
	for(i in 1:3){
		print(cor(2^(pure[,i]),deconv[,i]))
	}

chichaumiau/DSA documentation built on Dec. 19, 2021, 3:56 p.m.