R/CMprep.R

In bradnelms/CellMapper: Predict genes expressed selectively in specific cell types

CMprep <-
function(Data, DataSource = '', GeneIDType = '', verbose = TRUE) {

	if (is.matrix(Data) & is.numeric(Data)) {
		dataMat <- Data
		rm(Data)
	} else if (class(Data) == "ExpressionSet") {
		if (requireNamespace("Biobase")) {
			dataMat <- Biobase::exprs(Data)
			rm(Data)
		} else {
stop("'Biobase' must be installed to access an 'ExpressionSet' object")
		}
	} else {
		stop("'Data' must be either an Affy 'ExpressionSet' object or ",
		"a numeric matrix\n")
	}


	if (is.null(rownames(dataMat))) {
		stop("This expression matrix has no gene names, CellMapper ",
		"requires that the data matrix have gene names.")
	}

	if (any(is.na(dataMat))) {
		stop("There are NAs in the input data. CellMapper cannot ",
		"accept datasets with missing values. Consider using a ",
		"packages such as 'impute' to estimate the missing expression ",
		"values.")
	}

	if (!(is.character(DataSource) & is.character(GeneIDType) &
		(length(DataSource) == 1) & (length(GeneIDType) == 1))) {
		stop("'DataSource' and 'GeneIDType' must both be character ",
		"vectors of length = 1.")
	}

	if (verbose) {
		message("* Scaling the data matrix...\n")
	}
	dataMat <- sweep(dataMat, 1, apply(dataMat, 1, mean), "-")
	dataMat <- sweep(dataMat, 1, sqrt(apply(dataMat, 1, var)), "/")
		

	if (verbose) {
		message("* Computing the singular-value decomposition of the ",
		"data matrix...\n")
	}
	sv <- svd(dataMat)


	if (verbose) {
		message("* Preparing data for CellMapper...\n")
	}
	keep <- sum(length(sv$d)*sv$d/sum(sv$d) > 1)
	sv$u <- sv$u[,seq_len(keep)]
	sv$d = sv$d[seq_len(keep)]
	rownames(sv$u) <- rownames(dataMat)

	out <- CellMapperList(B = t(sv$u), d = sv$d, meta = list(nrow =
	dim(dataMat)[1], ncol = dim(dataMat)[2], GeneIDType = GeneIDType,
	DataSource = DataSource))

	return(out)
}