R/mixedSpeciesImput.R

In hkumar6/gene.expression: Imputing dropouts in gene expression

#' The mixedSpeciesImput class
#' 
#' This class has slots to store results of different simulations
#' using Lasso, Random Forest and kknn methods.
#' 
#' @slot simulation.result.genes a list storing results of all simulations
#'  where genes are imputed as function of cells
#' @slot simulation.result.cells a list storing results of all simulations
#'  where cells are imputed as functions of genes
#'  
#' @importFrom methods new
#' @exportClass mixedSpeciesImput

mixedSpeciesImput <- setClass("mixedSpeciesImput",
                              slots = 
                                c(
                                  simulation.result.genes = "list",
                                  simulation.result.cells = "list"
                                ),
                              prototype = list(
                                simulation.result.genes = list(),
                                simulation.result.cells = list()
                              ))


# method to run simulations on genes with dependency on cells
setGeneric(name = "mixedSpeciesGene",
           def = function(mixedSpeciesdata) {
             standardGeneric("mixedSpeciesGene")
           })

setMethod(f = "mixedSpeciesGene",
          signature = "ANY",
          definition = function(mixedSpeciesdata) {
            theObject = mixedSpeciesImput()
            #Reduce Number of Cells
            #Seperate between human and mouse genes
            human.simData <- mixedSpeciesdata[grep("HUMAN",rownames(mixedSpeciesdata)),]
            mouse.simData <- mixedSpeciesdata[grep("MOUSE",rownames(mixedSpeciesdata)),]
            simData <- rbind(human.simData,mouse.simData)
            
            #count number of gene expression along cells seperately for human and mouse genes
            human.counts <- colSums(human.simData != 0)
            mouse.counts <- colSums(mouse.simData != 0)
            
            #exclude cells that have a very similar count of gene expression in both species
            mixed.cells <- sort(abs(human.counts - mouse.counts))[1:20]
            mixed.cells <- names(mixed.cells)
            simData <- simData[,!(names(simData) %in% mixed.cells)]
            human.counts <- human.counts[!(names(human.counts) %in% mixed.cells)]
            mouse.counts <- mouse.counts[!(names(mouse.counts) %in% mixed.cells)]
            
            #seperate the data according to the counts
            human.cells <- colnames(simData)[which(human.counts > mouse.counts)]
            mouse.cells <- colnames(simData)[which(human.counts < mouse.counts)]
            human.cell.simData <- simData[,which(human.counts > mouse.counts)]
            mouse.cell.simData <- simData[,which(human.counts < mouse.counts)]
            
            #rename the cells starting with HUMAN or MOUSE 
            human.cells <- unlist(lapply(human.cells, function(x){ toString(c("HUMAN",x))}))
            mouse.cells <- unlist(lapply(mouse.cells, function(x){ toString(c("MOUSE",x))}))
            colnames(mouse.cell.simData) <- mouse.cells
            colnames(human.cell.simData) <- human.cells
            
            #select 100 most expressed cells from human and 100 from mouse
            #local.means.human <- data.frame(colMeans(human.cell.simData))
            #names(local.means.human) <- c("colMeans")
            #rownames(local.means.human)[order(local.means.human$colMeans, decreasing = TRUE)[1:100]] -> selectedCells.human
            #local.means.mouse <- data.frame(colMeans(mouse.cell.simData))
            #names(local.means.mouse) <- c("colMeans")
            #rownames(local.means.mouse)[order(local.means.mouse$colMeans, decreasing = TRUE)[1:100]] -> selectedCells.mouse
            
            #simData.human <- human.cell.simData[,selectedCells.human]
            #simData.mouse <- mouse.cell.simData[,selectedCells.mouse]
            
            #unite both data sets
            simData <- cbind(human.cell.simData, mouse.cell.simData)
            
            #Reduce Number of Genes (100 mouse and 100 human most expressed genes)
            local.means.human <- data.frame(rowMeans(human.simData))
            names(local.means.human) <- c("rowMeans")
            rownames(local.means.human)[order(local.means.human$rowMeans, decreasing = TRUE)[1:100]] -> selectedGenes.human
            local.means.mouse <- data.frame(rowMeans(mouse.simData))
            names(local.means.mouse) <- c("rowMeans")
            rownames(local.means.mouse)[order(local.means.mouse$rowMeans, decreasing = TRUE)[1:100]] -> selectedGenes.mouse
            
            simData.human <- simData[selectedGenes.human,]
            simData.mouse <- simData[selectedGenes.mouse,]
            
            simData <- rbind(simData.human,simData.mouse)

            # simulation for random forest
            rownames(simData) <- gsub("[_:-]", "", rownames(simData), perl = TRUE)
            td <- mapply(randomForest.mixed.data, rownames(simData),
                         MoreArgs = list(t(as.data.frame(simData))))
            theObject@simulation.result.genes[[length(theObject@simulation.result.genes)+1]] <- td
            attr(theObject@simulation.result.genes[[length(theObject@simulation.result.genes)]], "method") <- "randomForest"

            # simulation for lasso
            td <- mapply(lasso.mixed.data, rownames(simData),
                         MoreArgs = list(t(simData),genes = TRUE))
            theObject@simulation.result.genes[[length(theObject@simulation.result.genes)+1]] <- td
            attr(theObject@simulation.result.genes[[length(theObject@simulation.result.genes)]], "method") <- "lasso"
            
            # simulation for lasso Negative Binomial
            td <- mapply(lasso.negbin.mixed.data, rownames(simData),
                         MoreArgs = list(t(simData),genes = TRUE))
            theObject@simulation.result.genes[[length(theObject@simulation.result.genes)+1]] <- td
            attr(theObject@simulation.result.genes[[length(theObject@simulation.result.genes)]], "method") <- "lassoNegBin"
            
            # simulation for kknn
            rownames(simData) <- gsub("[_:-]", "", rownames(simData), perl = TRUE)
            td <- mapply(kknn.mixed.data, rownames(simData),
                         MoreArgs = list(t(as.data.frame(simData))))
            theObject@simulation.result.genes[[length(theObject@simulation.result.genes)+1]] <- td
            attr(theObject@simulation.result.genes[[length(theObject@simulation.result.genes)]], "method") <- "kknn"
            
            return(theObject)
          })

# method to run simulations on cells with dependency on genes
setGeneric(name = "mixedSpeciesCells",
           def = function(mixedSpeciesdata) {
             standardGeneric("mixedSpeciesCells")
           })

setMethod(f = "mixedSpeciesCells",
          signature = "ANY",
          definition = function(mixedSpeciesdata) {
            theObject = mixedSpeciesImput()
            #Reduce Number of Cells
            #Seperate between human and mouse genes
            human.simData <- mixedSpeciesdata[grep("HUMAN",rownames(mixedSpeciesdata)),]
            mouse.simData <- mixedSpeciesdata[grep("MOUSE",rownames(mixedSpeciesdata)),]
            simData <- rbind(human.simData,mouse.simData)
            
            #count number of gene expression along cells seperately for human and mouse genes
            human.counts <- colSums(human.simData != 0)
            mouse.counts <- colSums(mouse.simData != 0)
            
            #exclude cells that have a very similar count of gene expression in both species
            mixed.cells <- sort(abs(human.counts - mouse.counts))[1:20]
            mixed.cells <- names(mixed.cells)
            simData <- simData[,!(names(simData) %in% mixed.cells)]
            human.counts <- human.counts[!(names(human.counts) %in% mixed.cells)]
            mouse.counts <- mouse.counts[!(names(mouse.counts) %in% mixed.cells)]
            
            #seperate the data according to the counts
            human.cells <- colnames(simData)[which(human.counts > mouse.counts)]
            mouse.cells <- colnames(simData)[which(human.counts < mouse.counts)]
            human.cell.simData <- simData[,which(human.counts > mouse.counts)]
            mouse.cell.simData <- simData[,which(human.counts < mouse.counts)]
            
            #rename the cells starting with HUMAN or MOUSE 
            human.cells <- unlist(lapply(human.cells, function(x){ toString(c("HUMAN",x))}))
            mouse.cells <- unlist(lapply(mouse.cells, function(x){ toString(c("MOUSE",x))}))
            colnames(mouse.cell.simData) <- mouse.cells
            colnames(human.cell.simData) <- human.cells
            
            #select 100 most expressed cells from human and 100 from mouse
            #local.means.human <- data.frame(colMeans(human.cell.simData))
            #names(local.means.human) <- c("colMeans")
            #rownames(local.means.human)[order(local.means.human$colMeans, decreasing = TRUE)[1:100]] -> selectedCells.human
            #local.means.mouse <- data.frame(colMeans(mouse.cell.simData))
            #names(local.means.mouse) <- c("colMeans")
            #rownames(local.means.mouse)[order(local.means.mouse$colMeans, decreasing = TRUE)[1:100]] -> selectedCells.mouse
            
            #simData.human <- human.cell.simData[,selectedCells.human]
            #simData.mouse <- mouse.cell.simData[,selectedCells.mouse]
            
            #unite both data sets
            simData <- cbind(human.cell.simData, mouse.cell.simData)
            
            #Reduce Number of Genes (100 mouse and 100 human most expressed genes)
            local.means.human <- data.frame(rowMeans(human.simData))
            names(local.means.human) <- c("rowMeans")
            rownames(local.means.human)[order(local.means.human$rowMeans, decreasing = TRUE)[1:100]] -> selectedGenes.human
            local.means.mouse <- data.frame(rowMeans(mouse.simData))
            names(local.means.mouse) <- c("rowMeans")
            rownames(local.means.mouse)[order(local.means.mouse$rowMeans, decreasing = TRUE)[1:100]] -> selectedGenes.mouse
            
            simData.human <- simData[selectedGenes.human,]
            simData.mouse <- simData[selectedGenes.mouse,]
            
            simData <- rbind(simData.human,simData.mouse)
            
            # simulation for random forest
            td <- mapply(randomForest.mixed.data, colnames(simData),
                         MoreArgs = list(simData))
            theObject@simulation.result.cells[[length(theObject@simulation.result.cells)+1]] <- td
            attr(theObject@simulation.result.cells[[length(theObject@simulation.result.cells)]], "method") <- "randomForest"
            
            # simulation for lasso
            td <- mapply(lasso.mixed.data, colnames(simData),
                         MoreArgs = list(simData, genes = FALSE))
            theObject@simulation.result.cells[[length(theObject@simulation.result.cells)+1]] <- td
            attr(theObject@simulation.result.cells[[length(theObject@simulation.result.cells)]], "method") <- "lasso"
            
            # simulation for lasso Negative Binomial
            td <- mapply(lasso.negbin.mixed.data, colnames(simData),
                         MoreArgs = list(simData, genes = FALSE))
            theObject@simulation.result.cells[[length(theObject@simulation.result.cells)+1]] <- td
            attr(theObject@simulation.result.cells[[length(theObject@simulation.result.cells)]], "method") <- "lassoNegBin"
            
            
            # simulation for kknn
            colnames(simData) <- gsub("[ ,]", "", colnames(simData), perl = TRUE)
            td <- mapply(kknn.mixed.data, colnames(simData),
                         MoreArgs = list(as.data.frame(simData)))
            theObject@simulation.result.cells[[length(theObject@simulation.result.cells)+1]] <- td
            attr(theObject@simulation.result.cells[[length(theObject@simulation.result.cells)]], "method") <- "kknn"
            
            return(theObject)
          })