R/RCSL.R

In RCSL: Rank Constrained Similarity Learning for Single Cell RNA Sequencing Data

#'
#' Perform the RCSL program
#'
#' @param data normalizaed gene expression matrix(each column represents a cell)
#' @param GF should I need the gene filter step?
#' @param gfRatio the ratio of the gene filter
#' @param pcRatio the ratio between the variance of the choosed PCs and the total variance
#' @param pcRatio the ratio between the variance of the 
#’                 choosed PCs and the total variance
#' @param NN.method the method of finding neighbors
#' @param Dis.method the distance metric in finding neighbors                    
#' @param neiRatio ratio of the number of selected 
#‘                 neighbors to the total number of cells 
#' @return gfData gene expression matrix after genes filtering
#' @return B block-diagonal matrix
#' @return C estimated number of clusters
#' @return y clustering results
#' 
#' @export
#'
#' @examples 
#' data <- log2(yan[1:100,1:15]+1)
#' RCSL(data)

RCSL <- function(data, GF = TRUE, gfRatio = 0.025, pcRatio = 0.95,
                 NN.method = "KNN", Dis.method = "Euclidean", neiRatio = 0.65){

# Gene filter
if(GF == TRUE){
gfData <- GenesFilter(data, gfRatio)
} else{
gfData <- data[rowSums(data)!=0,,drop=FALSE]
}

# Calculate the similarity matrix S
resSimS <- SimS(gfData, pcRatio, NN.method = "KNN", Dis.method = "Euclidean")

# Estimate the number of clusters C
Estimated_C <- EstClusters(resSimS$drData,resSimS$S)

# Calculate the block diagonal matrix B
res_BDSM <- BDSM(resSimS$S, Estimated_C)

B <- res_BDSM$B
y <- res_BDSM$y

res <- list("gfData" = gfData, "drData" = resSimS$drData,
            "S" = resSimS$S, "B" = B, "C" = Estimated_C, "y" = y)
return(res)
}