R/runTSCAN.R
In compbiomed/singleCellTK: Comprehensive and Interactive Analysis of Single Cell RNA-Seq Data
Defines functions
.getClustersLineData
plotTSCANDimReduceFeatures
plotTSCANClusterDEG
plotTSCANClusterPseudo
runTSCANClusterDEAnalysis
plotTSCANPseudotimeGenes
.viridisPseudoTimeColor
plotTSCANPseudotimeHeatmap
runTSCANDEG
plotTSCANResults
runTSCAN
Documented in
plotTSCANClusterDEG
plotTSCANClusterPseudo
plotTSCANDimReduceFeatures
plotTSCANPseudotimeGenes
plotTSCANPseudotimeHeatmap
plotTSCANResults
runTSCAN
runTSCANClusterDEAnalysis
runTSCANDEG
###################################################
### Setting accessor functions
###################################################
#' @title getTSCANResults accessor function
#' @description SCTK allows user to access all TSCAN related results with
#' \code{"getTSCANResults"}. See details.
#' @param x Input \linkS4class{SingleCellExperiment} object.
#' @param analysisName Algorithm name implemented, should be one of
#' \code{"Pseudotime"}, \code{"DEG"}, or \code{"ClusterDEAnalysis"}.
#' @param pathName Sub folder name within the \code{analysisName}. See details.
#' @param value Value to be stored within the \code{pathName} or
#' \code{analysisName}
#' @details
#' When \code{analysisName = "Pseudotime"}, returns the list result from
#' \code{\link{runTSCAN}}, including the MST structure.
#'
#' When \code{analysisName = "DEG"}, returns the list result from
#' \code{\link{runTSCANDEG}}, including \code{DataFrame}s containing genes that
#' increase/decrease along each the pseudotime paths. \code{pathName} indicates
#' the path index, the available options of which can be listed by
#' \code{listTSCANTerminalNodes}.
#'
#' When \code{analysisName = "ClusterDEAnalysis"}, returns the list result from
#' \code{\link{runTSCANClusterDEAnalysis}}. Here \code{pathName} needs to match
#' with the \code{useCluster} argument when running the algorithm.
#' @export
#' @rdname getTSCANResults
#' @return Get or set TSCAN results
#' @examples
#' data("mouseBrainSubsetSCE", package = "singleCellTK")
#' mouseBrainSubsetSCE <- runTSCAN(inSCE = mouseBrainSubsetSCE,
#' useReducedDim = "PCA_logcounts")
#' results <- getTSCANResults(mouseBrainSubsetSCE, "Pseudotime")
setGeneric("getTSCANResults", signature = "x",
function(x, analysisName = NULL, pathName = NULL) {
standardGeneric("getTSCANResults")
}
)
#' @export
#' @rdname getTSCANResults
#'
setMethod("getTSCANResults", signature(x = "SingleCellExperiment"),
function(x, analysisName = NULL, pathName = NULL){
result.names <- listTSCANResults(x)
if(!analysisName %in% result.names) {
stop("The analysis was not found for TSCAN results")
}
if (analysisName == "Pseudotime")
results <- S4Vectors::metadata(x)$sctk$Traj$TSCAN$Pseudotime
else {
if (is.null(pathName)) {
results <- S4Vectors::metadata(x)$sctk$Traj$TSCAN[[analysisName]]
} else {
pathName <- as.character(pathName)
results <- S4Vectors::metadata(x)$sctk$Traj$TSCAN[[analysisName]][[pathName]]
}
}
return(results)
})
#' @export
#' @rdname getTSCANResults
#'
setGeneric("getTSCANResults<-",
function(x, analysisName, pathName = NULL, value)
standardGeneric("getTSCANResults<-")
)
#' @export
#' @rdname getTSCANResults
setReplaceMethod("getTSCANResults", signature(x = "SingleCellExperiment"),
function(x, analysisName, pathName = NULL, value) {
if (analysisName == "Pseudotime")
S4Vectors::metadata(x)$sctk$Traj$TSCAN$Pseudotime <- value
else {
if (is.null(pathName))
stop("Have to specify `pathName` for TSCAN DE analyses")
pathName <- as.character(pathName)
S4Vectors::metadata(x)$sctk$Traj$TSCAN[[analysisName]][[pathName]] <- value
}
return(x)
})
#' @export
#' @rdname getTSCANResults
setGeneric("listTSCANResults", signature = "x",
function(x) {standardGeneric("listTSCANResults")}
)
#' @export
#' @rdname getTSCANResults
setMethod("listTSCANResults", "SingleCellExperiment", function(x){
all.results <- S4Vectors::metadata(x)$sctk$Traj$TSCAN
if (is.null(all.results) ||
!"Pseudotime" %in% names(all.results)) {
stop("No TSCAN result found. Please run `runTSCAN` first.")
}
return(names(all.results))
})
#' @export
#' @rdname getTSCANResults
setGeneric("listTSCANTerminalNodes", signature = "x",
function(x) {
standardGeneric("listTSCANTerminalNodes")
}
)
#' @export
#' @rdname getTSCANResults
setMethod("listTSCANTerminalNodes", signature(x = "SingleCellExperiment"),
function(x){
if (is.null(S4Vectors::metadata(x)$sctk$Traj$TSCAN$Pseudotime))
stop("No TSCAN result found. Please run `runTSCAN()` first.")
results <- S4Vectors::metadata(x)$sctk$Traj$TSCAN$Pseudotime$pathClusters
return(names(results))
})
###################################################
### STEP 1:: creating cluster and MST
###################################################
#' @title Run TSCAN to obtain pseudotime values for cells
#' @description Wrapper for obtaining a pseudotime ordering of the cells by
#' projecting them onto the minimum spanning tree (MST)
#' @param inSCE Input \linkS4class{SingleCellExperiment} object.
#' @param useReducedDim Character. A low-dimension representation in
#' \code{reducedDims}, will be used for both clustering if \code{cluster} not
#' specified and MST construction. Default \code{"PCA"}.
#' @param cluster Grouping for each cell in \code{inSCE}. A vector with equal
#' length to the number of the cells in \code{inSCE}, or a single character for
#' retriving \code{colData} variable. Default \code{NULL}, will run
#' \code{runScranSNN} to obtain.
#' @param starter Character. Specifies the starting node from which to compute
#' the pseudotime. Default \code{NULL}, will select an arbitrary node.
#' @param seed An integer. Random seed for clustering if \code{cluster} is not
#' specified. Default \code{12345}.
#' @return The input \code{inSCE} object with pseudotime ordering of the cells
#' along the paths and the cluster label stored in \code{colData}, and other
#' unstructured information in \code{metadata}.
#' @export
#' @author Nida Pervaiz
#' @examples
#' data("mouseBrainSubsetSCE", package = "singleCellTK")
#' mouseBrainSubsetSCE <- runTSCAN(inSCE = mouseBrainSubsetSCE,
#' useReducedDim = "PCA_logcounts")
runTSCAN <- function(inSCE,
useReducedDim = "PCA",
cluster = NULL,
starter = NULL,
seed = 12345) {
if (is.null(cluster)) {
# DON'T RETURN TO `inSCE`
# It really overwrites existing cluster labels
sce <- runScranSNN(inSCE, useReducedDim = useReducedDim, clusterName = "scranSNN_cluster", k = 8, weightType = "jaccard", seed = seed)
cluster <- colData(sce)$scranSNN_cluster
rm(sce)
} else {
cluster <- .manageCellVar(inSCE, var = cluster)
}
if (length(unique(cluster)) == 1) {
stop("Only one cluster found. Unable to calculate.")
}
message(date(), " ... Running TSCAN to estimate pseudotime")
inSCE <- scran::computeSumFactors(inSCE, clusters = cluster)
by.cluster <- scuttle::aggregateAcrossCells(inSCE, ids = cluster)
centroids <- SingleCellExperiment::reducedDim(by.cluster, useReducedDim)
mst <- TSCAN::createClusterMST(centroids, clusters = NULL)
# Map each cell to the closest edge on the MST, reporting also the distance to
# the corresponding vertices.
map.tscan <- TSCAN::mapCellsToEdges(inSCE , mst = mst,
use.dimred = useReducedDim,
clusters = cluster)
# Compute a pseudotime for each cell lying on each path through the MST from a
# given starting node.
tscan.pseudo <- TSCAN::orderCells(map.tscan, mst, start = starter)
common.pseudo <- TrajectoryUtils::averagePseudotime(tscan.pseudo)
colData(inSCE)$TSCAN_clusters <- cluster
colData(inSCE)$TSCAN_pseudotime <- common.pseudo
pathClusters <- list()
pathList <- data.frame()
for(i in seq(ncol(tscan.pseudo))){
pathIndex = as.character(colnames(tscan.pseudo)[i])
colDataVarName <- paste0("Path_",pathIndex,"_pseudotime")
inSCE[[colDataVarName]] <- TrajectoryUtils::pathStat(tscan.pseudo)[,pathIndex]
nonNA.idx <- !is.na(colData(inSCE)[[colDataVarName]])
pathClusters[[pathIndex]] <- unique(colData(inSCE)$TSCAN_clusters[nonNA.idx])
pathList[i,1] <- pathIndex
pathList[i,2] <- toString(pathClusters[[pathIndex]])
message(date(), " ... Clusters involved in path index ", pathIndex,
" are: ", paste(pathClusters[[i]], collapse = ", "))
}
maxWidth <- max(stringr::str_length(pathList[, 1]))
choiceList <- paste0(stringr::str_pad(pathList[, 1], width=maxWidth,
side="right"),
"|", pathList[, 2])
branchClusters <- c()
edgeList <- mst
for (i in seq_along(unique(colData(inSCE)$TSCAN_clusters))){
degree <- sum(edgeList[i] != 0)
if (degree > 1) branchClusters <- c(branchClusters, i)
}
## Save these results in a list and then make S4 accessor that passes the
## entire list
result <- list(pseudo = tscan.pseudo, mst = mst,
maptscan = map.tscan,
pathClusters = pathClusters,
branchClusters = branchClusters,
pathIndexList = choiceList)
getTSCANResults(inSCE, analysisName = "Pseudotime") <- result
message(date(), " ... Number of estimated paths is ", ncol(tscan.pseudo))
return (inSCE)
}
###################################################
### plot pseudotime values
###################################################
#' @title Plot MST pseudotime values on cell 2D embedding
#' @description A wrapper function which visualizes outputs from the
#' \code{\link{runTSCAN}} function. Plots the pseudotime ordering of the cells
#' and project them onto the MST.
#' @param inSCE Input \linkS4class{SingleCellExperiment} object.
#' @param useReducedDim Saved dimension reduction name in \code{inSCE} object.
#' Required.
#' @return A \code{.ggplot} object with the pseudotime ordering of the cells
#' colored on a cell 2D embedding, and the MST path drawn on it.
#' @export
#' @author Nida Pervaiz
#' @examples
#' data("mouseBrainSubsetSCE", package = "singleCellTK")
#' mouseBrainSubsetSCE <- runTSCAN(inSCE = mouseBrainSubsetSCE,
#' useReducedDim = "PCA_logcounts")
#' plotTSCANResults(inSCE = mouseBrainSubsetSCE,
#' useReducedDim = "TSNE_logcounts")
plotTSCANResults <- function(inSCE, useReducedDim = "UMAP") {
results <- getTSCANResults(inSCE, analysisName = "Pseudotime")
clusters <- colData(inSCE)$TSCAN_clusters
by.cluster <- scuttle::aggregateAcrossCells(inSCE, ids = clusters)
line.data <- TSCAN::reportEdges(by.cluster, mst = results$mst,
clusters = NULL, use.dimred = useReducedDim)
scater::plotReducedDim(inSCE, dimred = useReducedDim,
colour_by = I(colData(inSCE)$TSCAN_pseudotime),
text_by = "TSCAN_clusters", text_colour = "black") +
suppressMessages(ggplot2::scale_colour_viridis_c(name = "Pseudotime")) +
ggplot2::geom_path(data = line.data,
ggplot2::aes_string(x = names(line.data)[2],
y = names(line.data)[3],
group = 'edge'))
}
###################################################
### STEP 2:: identify expressive genes
###################################################
#' @title Test gene expression changes along a TSCAN trajectory path
#' @description Wrapper for identifying genes with significant changes with
#' respect to one of the TSCAN pseudotime paths
#' @param inSCE Input \linkS4class{SingleCellExperiment} object.
#' @param pathIndex Path index for which the pseudotime values should be used.
#' This corresponds to the terminal node of specific path from the root
#' node to the terminal node. Run \code{listTSCANTerminalNodes(inSCE)} for
#' available options.
#' @param useAssay Character. The name of the assay to use for testing the
#' expression change. Should be log-normalized. Default \code{"logcounts"}
#' @param discardCluster Cluster(s) which are not of use or masks other
#' interesting effects can be discarded. Default \code{NULL}.
#' @return The input \code{inSCE} with results updated in \code{metadata}.
#' @export
#' @author Nida Pervaiz
#' @examples
#' data("mouseBrainSubsetSCE", package = "singleCellTK")
#' mouseBrainSubsetSCE <- runTSCAN(inSCE = mouseBrainSubsetSCE,
#' useReducedDim = "PCA_logcounts")
#' terminalNodes <- listTSCANTerminalNodes(mouseBrainSubsetSCE)
#' mouseBrainSubsetSCE <- runTSCANDEG(inSCE = mouseBrainSubsetSCE,
#' pathIndex = terminalNodes[1])
runTSCANDEG <- function(inSCE,
pathIndex,
useAssay = "logcounts",
discardCluster = NULL) {
nx <- inSCE
if (!is.null(discardCluster)) {
if (any(!discardCluster %in% unique(colData(nx)$TSCAN_clusters))) {
stop("Not all `discardCluster` exist in TSCAN clusters")
}
nx <- nx[, !(colData(nx)$TSCAN_clusters %in% discardCluster)]
}
pseudo <- TSCAN::testPseudotime(nx,
pseudotime = nx[[paste0("Path_", pathIndex,
"_pseudotime")]],
assay.type = useAssay)
pseudo <- pseudo[order(pseudo$FDR),]
up.left <- pseudo[pseudo$logFC < 0,]
up.right <- pseudo[pseudo$logFC > 0,]
## Save these results in a list and then make S4 accessor that passes the
## entire list
pathresults <- list(discardClusters = discardCluster, upLeft = up.left,
upRight = up.right, useAssay = useAssay)
getTSCANResults(inSCE, analysisName = "DEG",
pathName = as.character(pathIndex)) <- pathresults
return (inSCE)
}
###################################################
### plot heatmap of top genes
###################################################
#' @title Plot heatmap of genes with expression change along TSCAN pseudotime
#' @description A wrapper function which visualizes outputs from the
#' \code{\link{runTSCANDEG}} function. Plots the top genes that change in
#' expression with increasing pseudotime along the path in the MST.
#' \code{\link{runTSCANDEG}} has to be run in advance with using the same
#' \code{pathIndex} of interest.
#' @param inSCE Input \linkS4class{SingleCellExperiment} object.
#' @param pathIndex Path index for which the pseudotime values should be used.
#' Should have being used in \code{\link{runTSCANDEG}}.
#' @param direction Should we show features with expression increasing or
#' decreeasing along the increase in TSCAN pseudotime? Choices are
#' \code{"both"}, \code{"increasing"} or \code{"decreasing"}.
#' @param topN An integer. Only to plot this number of top genes along the path
#' in the MST, in terms of FDR value. Use \code{NULL} to cancel the top N
#' subscription. Default \code{30}.
#' @param log2fcThreshold Only output DEGs with the absolute values of log2FC
#' larger than this value. Default \code{NULL}.
#' @param useAssay A single character to specify a feature expression matrix in
#' \code{assays} slot. The expression of top features from here will be
#' visualized. Default \code{NULL} use the one used for
#' \code{\link{runTSCANDEG}}.
#' @param featureDisplay Whether to display feature ID and what ID type to
#' display. Users can set default ID type by \code{\link{setSCTKDisplayRow}}.
#' \code{NULL} will display when number of features to display is less than 60.
#' \code{FALSE} for no display. Variable name in \code{rowData} to indicate ID
#' type. \code{"rownames"} or \code{TRUE} for using \code{rownames(inSCE)}.
#' @return A ComplexHeatmap in \code{.ggplot} class
#' @export
#' @author Nida Pervaiz
#' @importFrom S4Vectors metadata
#' @examples
#' data("mouseBrainSubsetSCE", package = "singleCellTK")
#' mouseBrainSubsetSCE <- runTSCAN(inSCE = mouseBrainSubsetSCE,
#' useReducedDim = "PCA_logcounts")
#' terminalNodes <- listTSCANTerminalNodes(mouseBrainSubsetSCE)
#' mouseBrainSubsetSCE <- runTSCANDEG(inSCE = mouseBrainSubsetSCE,
#' pathIndex = terminalNodes[1])
#' plotTSCANPseudotimeHeatmap(mouseBrainSubsetSCE,
#' pathIndex = terminalNodes[1])
plotTSCANPseudotimeHeatmap <- function(inSCE,
pathIndex,
direction = c("both", "increasing",
"decreasing"),
topN = 50,
log2fcThreshold = NULL,
useAssay = NULL,
featureDisplay = metadata(inSCE)$featureDisplay){
results <- getTSCANResults(inSCE, analysisName = "DEG", pathName = pathIndex)
direction <- match.arg(direction)
# Organizing cells
cell.idx <- rep(TRUE, ncol(inSCE))
if(!is.null(results$discardClusters)){
cell.idx <- cell.idx &
(!colData(inSCE)$TSCAN_clusters %in% results$discardClusters)
}
colPathPseudo <- paste0("Path_", pathIndex, "_pseudotime")
cell.idx <- cell.idx & !is.na(colData(inSCE)[[colPathPseudo]])
cell.order <- order(colData(inSCE)[[colPathPseudo]])
cell.order <- cell.order[cell.order %in% which(cell.idx)]
# Organizing genes
if (direction == "both") {
degR <- results$upRight
degL <- results$upLeft
if (!is.null(log2fcThreshold)) {
degR <- degR[abs(degR$logFC) > log2fcThreshold,]
degL <- degL[abs(degL$logFC) > log2fcThreshold,]
}
genesR <- rownames(degR)[seq(min(topN, nrow(degR)))]
genesL <- rownames(degL)[seq(min(topN, nrow(degL)))]
genesR <- genesR[!is.na(genesR)]
genesL <- genesL[!is.na(genesL)]
genes <- c(genesL, genesR)
if (length(genes) < 1) stop("No feature passed the filter.")
direction.df <- data.frame(
Direction = factor(c(rep("decreasing", length(genesL)),
rep("increasing", length(genesR))),
levels = c("increasing", "decreasing")),
row.names = genes)
} else {
if (direction == "increasing") deg <- results$upRight
if (direction == "decreasing") deg <- results$upLeft
if (!is.null(log2fcThreshold)) {
deg <- deg[abs(deg$logFC) > log2fcThreshold,]
}
topN <- min(topN, nrow(deg))
if (topN < 1) stop("No feature passed the filter.")
genes <- rownames(deg)[seq(topN)]
direction.df <- data.frame(Direction = rep(direction, length(genes)),
row.names = genes)
}
rowLabel <- featureDisplay
if (is.null(featureDisplay)) {
if (length(genes) <= 60) rowLabel <- TRUE else rowLabel <- FALSE
} else if (featureDisplay == "rownames") {
rowLabel <- TRUE
}
cellAnnotationColor = list(
.viridisPseudoTimeColor(colData(inSCE)[[colPathPseudo]])
)
names(cellAnnotationColor) <- colPathPseudo
if (is.null(useAssay)) useAssay <- results$useAssay
plotSCEHeatmap(inSCE = inSCE,
useAssay = useAssay,
cellIndex = cell.order,
featureIndex = genes,
colDend = FALSE,
rowDend = FALSE,
cluster_columns = FALSE, cluster_rows = TRUE,
colDataName = c("TSCAN_clusters", colPathPseudo),
rowLabel = rowLabel,
featureAnnotations = direction.df,
rowSplitBy = "Direction",
cellAnnotationColor = cellAnnotationColor
)
}
#' Basing on pseudotime is presented within range from 0 to 100
#' Use the viridis color scale to match with `plotTSCANResult` embedding color
#' @param x numeric vector of pseudotime
#' @return function object returned by circlize::colorRamp2
#' @noRd
.viridisPseudoTimeColor <- function(x) {
a <- max(x, na.rm = TRUE)
b <- min(x, na.rm = TRUE)
chunk <- (a - b) / 4
circlize::colorRamp2(seq(0,4) * chunk + b,
c("#440154", "#3b528b", "#21918c", "#5ec962", "#fde725"))
}
###################################################
### plot expressive genes
###################################################
#' @title Plot expression changes of top features along a TSCAN pseudotime path
#' @description A wrapper function which visualizes outputs from the
#' \code{\link{runTSCANDEG}} function. Plots the genes that increase or decrease
#' in expression with increasing pseudotime along the path in the MST.
#' \code{\link{runTSCANDEG}} has to be run in advance with using the same
#' \code{pathIndex} of interest.
#' @param inSCE Input \linkS4class{SingleCellExperiment} object.
#' @param pathIndex Path index for which the pseudotime values should be used.
#' Should have being used in \code{\link{runTSCANDEG}}.
#' @param direction Should we show features with expression increasing or
#' decreeasing along the increase in TSCAN pseudotime? Choices are
#' \code{"increasing"} or \code{"decreasing"}.
#' @param topN An integer. Only to plot this number of top genes that are
#' increasing/decreasing in expression with increasing pseudotime along
#' the path in the MST. Default 10
#' @param useAssay A single character to specify a feature expression matrix in
#' \code{assays} slot. The expression of top features from here will be
#' visualized. Default \code{NULL} use the one used for
#' \code{\link{runTSCANDEG}}.
#' @param featureDisplay Specify the feature ID type to display. Users can set
#' default value with \code{\link{setSCTKDisplayRow}}. \code{NULL} or
#' \code{"rownames"} specifies the rownames of \code{inSCE}. Other character
#' values indicates \code{rowData} variable.
#' @return A \code{.ggplot} object with the facets of the top genes. Expression
#' on y-axis, pseudotime on x-axis.
#' @export
#' @author Nida Pervaiz
#' @importFrom utils head
#' @importFrom S4Vectors metadata
#' @importFrom SummarizedExperiment rowData
#' @examples
#' data("mouseBrainSubsetSCE", package = "singleCellTK")
#' mouseBrainSubsetSCE <- runTSCAN(inSCE = mouseBrainSubsetSCE,
#' useReducedDim = "PCA_logcounts")
#' terminalNodes <- listTSCANTerminalNodes(mouseBrainSubsetSCE)
#' mouseBrainSubsetSCE <- runTSCANDEG(inSCE = mouseBrainSubsetSCE,
#' pathIndex = terminalNodes[1])
#' plotTSCANPseudotimeGenes(mouseBrainSubsetSCE,
#' pathIndex = terminalNodes[1],
#' useAssay = "logcounts")
plotTSCANPseudotimeGenes <- function(inSCE,
pathIndex,
direction = c("increasing", "decreasing"),
topN = 10,
useAssay = NULL,
featureDisplay = metadata(inSCE)$featureDisplay){
results <- getTSCANResults(inSCE, analysisName = "DEG", pathName = pathIndex)
direction = match.arg(direction)
if (!is.null(featureDisplay) &&
featureDisplay == "rownames")
featureDisplay <- NULL
if(!is.null(results$discardClusters)){
inSCE <- inSCE[,!colData(inSCE)$TSCAN_clusters %in% results$discardClusters]
}
if (direction == "decreasing") features <- rownames(results$upLeft)
if (direction == "increasing") features <- rownames(results$upRight)
features <- head(features, topN)
if (!is.null(featureDisplay)) {
features.idx <- featureIndex(features, inSCE)
features <- rowData(inSCE)[[featureDisplay]][features.idx]
}
if (is.null(useAssay)) useAssay <- results$useAssay
scater::plotExpression(inSCE,
features = features,
exprs_values = useAssay,
swap_rownames = featureDisplay,
x = paste0("Path_",pathIndex,"_pseudotime"),
colour_by = "TSCAN_clusters")
}
###################################################
### STEP3:: identify DE genes in branch cluster
###################################################
#' @title Find DE genes between all TSCAN paths rooted from given cluster
#' @description This function finds all paths that root from a given cluster
#' \code{useCluster}, and performs tests to identify significant features for
#' each path, and are not significant and/or changing in the opposite direction
#' in the other paths. Using a branching cluster (i.e. a node with degree > 2)
#' may highlight features which are responsible for the branching event. MST has
#' to be pre-calculated with \code{\link{runTSCAN}}.
#' @param inSCE Input \linkS4class{SingleCellExperiment} object.
#' @param useCluster The cluster to be regarded as the root, has to existing in
#' \code{colData(inSCE)$TSCAN_clusters}.
#' @param useAssay Character. The name of the assay to use. This assay should
#' contain log normalized counts. Default \code{"logcounts"}.
#' @param fdrThreshold Only out put DEGs with FDR value smaller than this value.
#' Default \code{0.05}.
#' @return The input \code{inSCE} with results updated in \code{metadata}.
#' @export
#' @author Nida Pervaiz
#' @examples
#' data("mouseBrainSubsetSCE", package = "singleCellTK")
#' mouseBrainSubsetSCE <- runTSCAN(inSCE = mouseBrainSubsetSCE,
#' useReducedDim = "PCA_logcounts")
#' mouseBrainSubsetSCE <- runTSCANClusterDEAnalysis(inSCE = mouseBrainSubsetSCE,
#' useCluster = 1)
runTSCANClusterDEAnalysis <- function(inSCE,
useCluster,
useAssay = "logcounts",
fdrThreshold = 0.05) {
results <- getTSCANResults(inSCE, analysisName = "Pseudotime")
message(date(), " ... Finding DEG between TSCAN branches")
tscan.pseudo <- TSCAN::orderCells(results$maptscan, mst = results$mst,
start = useCluster)
nPaths <- colnames(tscan.pseudo)
pathClusters <- list()
pathList <- data.frame()
x <- inSCE[,colData(inSCE)$TSCAN_clusters == useCluster]
store <- list()
genes <- list()
for(i in seq(nPaths)){
# Get Pseudotime of a path
pathIndex = as.character(nPaths[i])
branchPseudotime <- TrajectoryUtils::pathStat(tscan.pseudo)[,pathIndex]
colDataVarName <- paste0("TSCAN_Cluster", useCluster,
"_Path_", pathIndex, "_pseudotime")
colData(inSCE)[[colDataVarName]] <- branchPseudotime
nonNA.idx <- !is.na(branchPseudotime)
pathClusters[[pathIndex]] <- unique(colData(inSCE)$TSCAN_clusters[nonNA.idx])
pathList[i,1] <- pathIndex
pathList[i,2] <- toString(pathClusters[[pathIndex]])
message(date(), " ... Clusters involved in path index ", pathIndex,
" are: ", paste(pathClusters[[i]], collapse = ", "))
# Test along the path
pseudo.df <- TSCAN::testPseudotime(
x, df = 1,
pseudotime = branchPseudotime[colData(inSCE)$TSCAN_clusters == useCluster],
assay.type = useAssay)
pseudo.df <- pseudo.df[order(pseudo.df$p.value),]
store[[pathIndex]] <- pseudo.df
}
# Filter against all other paths
genes <- list()
for(i in seq(nPaths)){
pseudo.df <- store[[nPaths[i]]]
paths <- setdiff(seq(nPaths), i)
idx <- pseudo.df$FDR < fdrThreshold
for(j in seq_along(paths)){
pseudo.otherPath <- store[[paths[j]]]
idx <- idx & (pseudo.otherPath$p.value >= 0.05 |
sign(pseudo.otherPath$logFC) != sign(pseudo.df$logFC))
}
pseudo.df <- pseudo.df[which(idx), ]
genes[[nPaths[i]]] <- pseudo.df[order(pseudo.df$FDR),]
}
maxWidth <- max(stringr::str_length(pathList[, 1]))
choiceList <- paste0(stringr::str_pad(pathList[, 1], width=maxWidth,
side="right"),
"|", pathList[, 2])
expGenes <- list(DEgenes = genes, useAssay = useAssay,
terminalNodes = tscan.pseudo,
pathIndexList = choiceList)
getTSCANResults(inSCE, analysisName = "ClusterDEAnalysis",
pathName = useCluster) <- expGenes
return(inSCE)
}
###################################################
### plot branch cluster
###################################################
#' @title Plot TSCAN pseudotime rooted from given cluster
#' @description This function finds all paths that root from a given cluster
#' \code{useCluster}. For each path, this function plots the recomputed
#' pseudotime starting from the root on a scatter plot which contains cells only
#' in this cluster. MST has to be pre-calculated with \code{\link{runTSCAN}}.
#' @param inSCE Input \linkS4class{SingleCellExperiment} object.
#' @param useCluster The cluster to be regarded as the root, has to existing in
#' \code{colData(inSCE)$TSCAN_clusters}.
#' @param useReducedDim Saved dimension reduction name in the
#' SingleCellExperiment object. Required.
#' @param combinePlot Must be either \code{"all"} or \code{"none"}. \code{"all"}
#' will combine plots of pseudotime along each path into a single \code{.ggplot}
#' object, while \code{"none"} will output a list of plots. Default
#' \code{"all"}.
#' @return
#' \item{combinePlot = "all"}{A \code{.ggplot} object}
#' \item{combinePlot = "none"}{A list of \code{.ggplot}}
#' @export
#' @author Nida Pervaiz
#' @importFrom utils head
#' @examples
#' data("mouseBrainSubsetSCE", package = "singleCellTK")
#' mouseBrainSubsetSCE <- runTSCAN(inSCE = mouseBrainSubsetSCE,
#' useReducedDim = "PCA_logcounts")
#' plotTSCANClusterPseudo(mouseBrainSubsetSCE, useCluster = 1,
#' useReducedDim = "TSNE_logcounts")
plotTSCANClusterPseudo <- function(inSCE, useCluster, useReducedDim = "UMAP",
combinePlot = c("all", "none")){
# Get the plotting info of edges
results <- getTSCANResults(inSCE, analysisName = "Pseudotime")
combinePlot <- match.arg(combinePlot)
clusters <- colData(inSCE)$TSCAN_clusters
by.cluster <- scuttle::aggregateAcrossCells(inSCE, ids = clusters)
line.data <- TSCAN::reportEdges(by.cluster, mst = results$mst,
clusters = NULL, use.dimred = useReducedDim)
line.data.sub <- .getClustersLineData(line.data, useCluster)
# Get Branch pseudotime
tscan.pseudo <- TSCAN::orderCells(results$maptscan, mst = results$mst,
start = useCluster)
nPaths <- colnames(tscan.pseudo)
x <- inSCE[,colData(inSCE)$TSCAN_clusters == useCluster]
plotList <- list()
for (i in seq(nPaths)) {
branchPseudotime <- TrajectoryUtils::pathStat(tscan.pseudo)[,i]
branchPseudotime <- branchPseudotime[colData(inSCE)$TSCAN_clusters == useCluster]
x$pseudotime <- branchPseudotime
plotList[[i]] <- scater::plotReducedDim(x, dimred = useReducedDim,
colour_by = "pseudotime",
text_by = "TSCAN_clusters",
text_colour = "black") +
ggplot2::geom_line(data = line.data.sub,
ggplot2::aes_string(x = colnames(line.data.sub)[2],
y = colnames(line.data.sub)[3],
group = "edge"))
}
if (combinePlot == "all") {
plotList <- cowplot::plot_grid(plotlist = plotList)
}
return(plotList)
}
###################################################
### plot gene of interest in branch cluster
###################################################
#' @title Plot features identified by \code{\link{runTSCANClusterDEAnalysis}} on
#' cell 2D embedding with MST overlaid
#' @description A wrapper function which plot the top features expression
#' identified by \code{\link{runTSCANClusterDEAnalysis}} on the 2D embedding of
#' the cells cluster used in the analysis. The related MST edges are overlaid.
#' @param inSCE Input \linkS4class{SingleCellExperiment} object.
#' @param useCluster Choose a cluster used for identifying DEG with
#' \code{\link{runTSCANClusterDEAnalysis}}. Required.
#' @param pathIndex Specifies one of the branching paths from \code{useCluster}
#' and plot the top DEGs on this path. Ususally presented by the terminal
#' cluster of a path. By default \code{NULL} plot top DEGs of all paths.
#' @param useReducedDim A single character for the matrix of 2D embedding.
#' Should exist in \code{reducedDims} slot. Default \code{"UMAP"}.
#' @param topN Integer. Use top N genes identified. Default \code{9}.
#' @param useAssay A single character for the feature expression matrix. Should
#' exist in \code{assayNames(inSCE)}. Default \code{NULL} for using the one used
#' in \code{\link{runTSCANClusterDEAnalysis}}.
#' @param featureDisplay Specify the feature ID type to display. Users can set
#' default value with \code{\link{setSCTKDisplayRow}}. \code{NULL} or
#' \code{"rownames"} specifies the rownames of \code{inSCE}. Other character
#' values indicates \code{rowData} variable.
#' @param combinePlot Must be either \code{"all"} or \code{"none"}. \code{"all"}
#' will combine plots of each feature into a single \code{.ggplot} object,
#' while \code{"none"} will output a list of plots. Default \code{"all"}.
#' @return A \code{.ggplot} object of cell scatter plot, colored by the
#' expression of a gene identified by \code{\link{runTSCANClusterDEAnalysis}},
#' with the layer of trajectory.
#' @export
#' @author Yichen Wang
#' @importFrom S4Vectors metadata
#' @examples
#' data("mouseBrainSubsetSCE", package = "singleCellTK")
#' mouseBrainSubsetSCE <- runTSCAN(inSCE = mouseBrainSubsetSCE,
#' useReducedDim = "PCA_logcounts")
#' mouseBrainSubsetSCE <- runTSCANClusterDEAnalysis(inSCE = mouseBrainSubsetSCE,
#' useCluster = 1)
#' plotTSCANClusterDEG(mouseBrainSubsetSCE, useCluster = 1,
#' useReducedDim = "TSNE_logcounts")
plotTSCANClusterDEG <- function(
inSCE,
useCluster,
pathIndex = NULL,
useReducedDim = "UMAP",
topN = 9,
useAssay = NULL,
featureDisplay = metadata(inSCE)$featureDisplay,
combinePlot = c("all", "none")
) {
MSTResults <- getTSCANResults(inSCE, analysisName = "Pseudotime")
if (length(useCluster) != 1) stop("Can only specify one cluster")
DEResults <- getTSCANResults(inSCE, analysisName = "ClusterDEAnalysis",
pathName = useCluster)
if (is.null(DEResults)) {
stop("Branch DE result of specified cluster not found. Run ",
"`runTSCANClusterDEAnalysis()` with the same `useCluster` first.")
}
combinePlot <- match.arg(combinePlot)
if (is.null(pathIndex)) {
deg <- do.call(rbind, DEResults$DEgenes)
deg <- deg[order(deg$FDR),]
} else {
pathIndex <- as.character(pathIndex)
deg <- DEResults$DEgenes[[pathIndex]]
}
features <- rownames(deg)[seq(min(topN, nrow(deg), na.rm = TRUE))]
if (is.null(useAssay)) useAssay <- DEResults$useAssay
plotTSCANDimReduceFeatures(inSCE = inSCE, features = features,
useReducedDim = useReducedDim,
useAssay = useAssay, useCluster = useCluster,
featureDisplay = featureDisplay,
combinePlot = combinePlot)
}
#' @title Plot feature expression on cell 2D embedding with MST overlaid
#' @description A wrapper function which plots all cells or cells in chosen
#' cluster. Each point is a cell colored by the expression of a feature of
#' interest, the relevant edges of the MST are overlaid on top.
#' @param inSCE Input \linkS4class{SingleCellExperiment} object.
#' @param features Choose the feature of interest to explore the expression
#' level on the trajectory. Required.
#' @param useReducedDim A single character for the matrix of 2D embedding.
#' Should exist in \code{reducedDims} slot. Default \code{"UMAP"}.
#' @param useAssay A single character for the feature expression matrix. Should
#' exist in \code{assayNames(inSCE)}. Default \code{"logcounts"}.
#' @param by Where should \code{features} be found? \code{NULL},
#' \code{"rownames"} for \code{rownames(inSCE)}, otherwise will be regarded as
#' \code{rowData} variable.
#' @param useCluster Choose specific clusters where gene expression needs to be
#' visualized. By default \code{NULL}, all clusters are chosen.
#' @param featureDisplay Specify the feature ID type to display. Users can set
#' default value with \code{\link{setSCTKDisplayRow}}. \code{NULL} or
#' \code{"rownames"} specifies the rownames of \code{inSCE}. Other character
#' values indicates \code{rowData} variable.
#' @param combinePlot Must be either \code{"all"} or \code{"none"}. \code{"all"}
#' will combine plots of each feature into a single \code{.ggplot} object,
#' while \code{"none"} will output a list of plots. Default \code{"all"}.
#' @return A \code{.ggplot} object of cell scatter plot, colored by the
#' expression of a gene of interest, with the layer of trajectory.
#' @export
#' @author Yichen Wang
#' @importFrom S4Vectors metadata
#' @examples
#' data("mouseBrainSubsetSCE", package = "singleCellTK")
#' mouseBrainSubsetSCE <- runTSCAN(inSCE = mouseBrainSubsetSCE,
#' useReducedDim = "PCA_logcounts")
#' plotTSCANDimReduceFeatures(inSCE = mouseBrainSubsetSCE,
#' features = "Tshz1",
#' useReducedDim = "TSNE_logcounts")
plotTSCANDimReduceFeatures <- function(
inSCE,
features,
useReducedDim = "UMAP",
useAssay = "logcounts",
by = "rownames",
useCluster = NULL,
featureDisplay = metadata(inSCE)$featureDisplay,
combinePlot = c("all", "none"))
{
results <- getTSCANResults(inSCE, analysisName = "Pseudotime")
combinePlot <- match.arg(combinePlot)
clusters <- colData(inSCE)$TSCAN_clusters
by.cluster <- scuttle::aggregateAcrossCells(inSCE, ids = clusters)
line.data <- TSCAN::reportEdges(by.cluster, mst = results$mst,
clusters = NULL, use.dimred = useReducedDim)
if (!is.null(useCluster)) {
line.data <- .getClustersLineData(line.data, useCluster)
inSCE <- inSCE[, clusters %in% useCluster]
}
if (by == "rownames") by <- NULL
plotList <- list()
features <- stats::na.omit(features)
for (f in features) {
# Should not enter the loop if features is length zero after NA omit
g <- plotSCEDimReduceFeatures(inSCE, feature = f,
useAssay = useAssay,
featureLocation = by,dim1 = 1, dim2 = 2,
featureDisplay = featureDisplay,
reducedDimName = useReducedDim,
title = f) +
ggplot2::geom_line(data = line.data,
ggplot2::aes_string(x = colnames(line.data)[2],
y = colnames(line.data)[3],
group = "edge"),
inherit.aes = FALSE)
# Text labeling code credit: scater
reddim <- as.data.frame(SingleCellExperiment::reducedDim(inSCE,
useReducedDim))
text_out <- scater::retrieveCellInfo(inSCE, "TSCAN_clusters",
search = "colData")
by_text_x <- vapply(split(reddim[,1], text_out$val),
stats::median, FUN.VALUE = 0)
by_text_y <- vapply(split(reddim[,2], text_out$val),
stats::median, FUN.VALUE = 0)
g <- g + ggrepel::geom_text_repel(
data = data.frame(x = by_text_x,
y = by_text_y,
label = names(by_text_x)),
mapping = ggplot2::aes_string(x = "x",
y = "y",
label = "label"),
inherit.aes = FALSE, na.rm = TRUE,
)
plotList[[f]] <- g
}
if (combinePlot == "all") {
if (length(plotList) > 0)
plotList <- cowplot::plot_grid(plotlist = plotList)
}
return(plotList)
}
#' Extract edges that connects to the clusters of interest
#' @param line.data data.frame of three columns. First column should be named
#' "edge", the other two are coordinates of one of the vertices that this
#' edge connects.
#' @param useCluster Vector of clusters, that are going to be the vertices for
#' edge finding.
#' @return data.frame, subset rows of line.data
#' @noRd
.getClustersLineData <- function(line.data, useCluster) {
idx <- vapply(useCluster, function(x){
grepl(paste0("^", x, "--"), line.data$edge) |
grepl(paste0("--", x, "$"), line.data$edge)
}, logical(nrow(line.data)))
# `idx` returned was c("matrix", "array") class. Each column is a logical
# vector for edge selection of each cluster.
# Next, do "or" for each row.
idx <- rowSums(idx) > 0
line.data[idx,]
}
compbiomed/singleCellTK documentation built on May 8, 2024, 6:58 p.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Defines functions .getClustersLineData plotTSCANDimReduceFeatures plotTSCANClusterDEG plotTSCANClusterPseudo runTSCANClusterDEAnalysis plotTSCANPseudotimeGenes .viridisPseudoTimeColor plotTSCANPseudotimeHeatmap runTSCANDEG plotTSCANResults runTSCAN
Documented in plotTSCANClusterDEG plotTSCANClusterPseudo plotTSCANDimReduceFeatures plotTSCANPseudotimeGenes plotTSCANPseudotimeHeatmap plotTSCANResults runTSCAN runTSCANClusterDEAnalysis runTSCANDEG
###################################################
### Setting accessor functions
###################################################
#' @title getTSCANResults accessor function
#' @description SCTK allows user to access all TSCAN related results with
#' \code{"getTSCANResults"}. See details.
#' @param x Input \linkS4class{SingleCellExperiment} object.
#' @param analysisName Algorithm name implemented, should be one of
#' \code{"Pseudotime"}, \code{"DEG"}, or \code{"ClusterDEAnalysis"}.
#' @param pathName Sub folder name within the \code{analysisName}. See details.
#' @param value Value to be stored within the \code{pathName} or
#' \code{analysisName}
#' @details
#' When \code{analysisName = "Pseudotime"}, returns the list result from
#' \code{\link{runTSCAN}}, including the MST structure.
#'
#' When \code{analysisName = "DEG"}, returns the list result from
#' \code{\link{runTSCANDEG}}, including \code{DataFrame}s containing genes that
#' increase/decrease along each the pseudotime paths. \code{pathName} indicates
#' the path index, the available options of which can be listed by
#' \code{listTSCANTerminalNodes}.
#'
#' When \code{analysisName = "ClusterDEAnalysis"}, returns the list result from
#' \code{\link{runTSCANClusterDEAnalysis}}. Here \code{pathName} needs to match
#' with the \code{useCluster} argument when running the algorithm.
#' @export
#' @rdname getTSCANResults
#' @return Get or set TSCAN results
#' @examples
#' data("mouseBrainSubsetSCE", package = "singleCellTK")
#' mouseBrainSubsetSCE <- runTSCAN(inSCE = mouseBrainSubsetSCE,
#' useReducedDim = "PCA_logcounts")
#' results <- getTSCANResults(mouseBrainSubsetSCE, "Pseudotime")
setGeneric("getTSCANResults", signature = "x",
function(x, analysisName = NULL, pathName = NULL) {
standardGeneric("getTSCANResults")
}
)
#' @export
#' @rdname getTSCANResults
#'
setMethod("getTSCANResults", signature(x = "SingleCellExperiment"),
function(x, analysisName = NULL, pathName = NULL){
result.names <- listTSCANResults(x)
if(!analysisName %in% result.names) {
stop("The analysis was not found for TSCAN results")
}
if (analysisName == "Pseudotime")
results <- S4Vectors::metadata(x)$sctk$Traj$TSCAN$Pseudotime
else {
if (is.null(pathName)) {
results <- S4Vectors::metadata(x)$sctk$Traj$TSCAN[[analysisName]]
} else {
pathName <- as.character(pathName)
results <- S4Vectors::metadata(x)$sctk$Traj$TSCAN[[analysisName]][[pathName]]
}
}
return(results)
})
#' @export
#' @rdname getTSCANResults
#'
setGeneric("getTSCANResults<-",
function(x, analysisName, pathName = NULL, value)
standardGeneric("getTSCANResults<-")
)
#' @export
#' @rdname getTSCANResults
setReplaceMethod("getTSCANResults", signature(x = "SingleCellExperiment"),
function(x, analysisName, pathName = NULL, value) {
if (analysisName == "Pseudotime")
S4Vectors::metadata(x)$sctk$Traj$TSCAN$Pseudotime <- value
else {
if (is.null(pathName))
stop("Have to specify `pathName` for TSCAN DE analyses")
pathName <- as.character(pathName)
S4Vectors::metadata(x)$sctk$Traj$TSCAN[[analysisName]][[pathName]] <- value
}
return(x)
})
#' @export
#' @rdname getTSCANResults
setGeneric("listTSCANResults", signature = "x",
function(x) {standardGeneric("listTSCANResults")}
)
#' @export
#' @rdname getTSCANResults
setMethod("listTSCANResults", "SingleCellExperiment", function(x){
all.results <- S4Vectors::metadata(x)$sctk$Traj$TSCAN
if (is.null(all.results) ||
!"Pseudotime" %in% names(all.results)) {
stop("No TSCAN result found. Please run `runTSCAN` first.")
}
return(names(all.results))
})
#' @export
#' @rdname getTSCANResults
setGeneric("listTSCANTerminalNodes", signature = "x",
function(x) {
standardGeneric("listTSCANTerminalNodes")
}
)
#' @export
#' @rdname getTSCANResults
setMethod("listTSCANTerminalNodes", signature(x = "SingleCellExperiment"),
function(x){
if (is.null(S4Vectors::metadata(x)$sctk$Traj$TSCAN$Pseudotime))
stop("No TSCAN result found. Please run `runTSCAN()` first.")
results <- S4Vectors::metadata(x)$sctk$Traj$TSCAN$Pseudotime$pathClusters
return(names(results))
})
###################################################
### STEP 1:: creating cluster and MST
###################################################
#' @title Run TSCAN to obtain pseudotime values for cells
#' @description Wrapper for obtaining a pseudotime ordering of the cells by
#' projecting them onto the minimum spanning tree (MST)
#' @param inSCE Input \linkS4class{SingleCellExperiment} object.
#' @param useReducedDim Character. A low-dimension representation in
#' \code{reducedDims}, will be used for both clustering if \code{cluster} not
#' specified and MST construction. Default \code{"PCA"}.
#' @param cluster Grouping for each cell in \code{inSCE}. A vector with equal
#' length to the number of the cells in \code{inSCE}, or a single character for
#' retriving \code{colData} variable. Default \code{NULL}, will run
#' \code{runScranSNN} to obtain.
#' @param starter Character. Specifies the starting node from which to compute
#' the pseudotime. Default \code{NULL}, will select an arbitrary node.
#' @param seed An integer. Random seed for clustering if \code{cluster} is not
#' specified. Default \code{12345}.
#' @return The input \code{inSCE} object with pseudotime ordering of the cells
#' along the paths and the cluster label stored in \code{colData}, and other
#' unstructured information in \code{metadata}.
#' @export
#' @author Nida Pervaiz
#' @examples
#' data("mouseBrainSubsetSCE", package = "singleCellTK")
#' mouseBrainSubsetSCE <- runTSCAN(inSCE = mouseBrainSubsetSCE,
#' useReducedDim = "PCA_logcounts")
runTSCAN <- function(inSCE,
useReducedDim = "PCA",
cluster = NULL,
starter = NULL,
seed = 12345) {
if (is.null(cluster)) {
# DON'T RETURN TO `inSCE`
# It really overwrites existing cluster labels
sce <- runScranSNN(inSCE, useReducedDim = useReducedDim, clusterName = "scranSNN_cluster", k = 8, weightType = "jaccard", seed = seed)
cluster <- colData(sce)$scranSNN_cluster
rm(sce)
} else {
cluster <- .manageCellVar(inSCE, var = cluster)
}
if (length(unique(cluster)) == 1) {
stop("Only one cluster found. Unable to calculate.")
}
message(date(), " ... Running TSCAN to estimate pseudotime")
inSCE <- scran::computeSumFactors(inSCE, clusters = cluster)
by.cluster <- scuttle::aggregateAcrossCells(inSCE, ids = cluster)
centroids <- SingleCellExperiment::reducedDim(by.cluster, useReducedDim)
mst <- TSCAN::createClusterMST(centroids, clusters = NULL)
# Map each cell to the closest edge on the MST, reporting also the distance to
# the corresponding vertices.
map.tscan <- TSCAN::mapCellsToEdges(inSCE , mst = mst,
use.dimred = useReducedDim,
clusters = cluster)
# Compute a pseudotime for each cell lying on each path through the MST from a
# given starting node.
tscan.pseudo <- TSCAN::orderCells(map.tscan, mst, start = starter)
common.pseudo <- TrajectoryUtils::averagePseudotime(tscan.pseudo)
colData(inSCE)$TSCAN_clusters <- cluster
colData(inSCE)$TSCAN_pseudotime <- common.pseudo
pathClusters <- list()
pathList <- data.frame()
for(i in seq(ncol(tscan.pseudo))){
pathIndex = as.character(colnames(tscan.pseudo)[i])
colDataVarName <- paste0("Path_",pathIndex,"_pseudotime")
inSCE[[colDataVarName]] <- TrajectoryUtils::pathStat(tscan.pseudo)[,pathIndex]
nonNA.idx <- !is.na(colData(inSCE)[[colDataVarName]])
pathClusters[[pathIndex]] <- unique(colData(inSCE)$TSCAN_clusters[nonNA.idx])
pathList[i,1] <- pathIndex
pathList[i,2] <- toString(pathClusters[[pathIndex]])
message(date(), " ... Clusters involved in path index ", pathIndex,
" are: ", paste(pathClusters[[i]], collapse = ", "))
}
maxWidth <- max(stringr::str_length(pathList[, 1]))
choiceList <- paste0(stringr::str_pad(pathList[, 1], width=maxWidth,
side="right"),
"|", pathList[, 2])
branchClusters <- c()
edgeList <- mst
for (i in seq_along(unique(colData(inSCE)$TSCAN_clusters))){
degree <- sum(edgeList[i] != 0)
if (degree > 1) branchClusters <- c(branchClusters, i)
}
## Save these results in a list and then make S4 accessor that passes the
## entire list
result <- list(pseudo = tscan.pseudo, mst = mst,
maptscan = map.tscan,
pathClusters = pathClusters,
branchClusters = branchClusters,
pathIndexList = choiceList)
getTSCANResults(inSCE, analysisName = "Pseudotime") <- result
message(date(), " ... Number of estimated paths is ", ncol(tscan.pseudo))
return (inSCE)
}
###################################################
### plot pseudotime values
###################################################
#' @title Plot MST pseudotime values on cell 2D embedding
#' @description A wrapper function which visualizes outputs from the
#' \code{\link{runTSCAN}} function. Plots the pseudotime ordering of the cells
#' and project them onto the MST.
#' @param inSCE Input \linkS4class{SingleCellExperiment} object.
#' @param useReducedDim Saved dimension reduction name in \code{inSCE} object.
#' Required.
#' @return A \code{.ggplot} object with the pseudotime ordering of the cells
#' colored on a cell 2D embedding, and the MST path drawn on it.
#' @export
#' @author Nida Pervaiz
#' @examples
#' data("mouseBrainSubsetSCE", package = "singleCellTK")
#' mouseBrainSubsetSCE <- runTSCAN(inSCE = mouseBrainSubsetSCE,
#' useReducedDim = "PCA_logcounts")
#' plotTSCANResults(inSCE = mouseBrainSubsetSCE,
#' useReducedDim = "TSNE_logcounts")
plotTSCANResults <- function(inSCE, useReducedDim = "UMAP") {
results <- getTSCANResults(inSCE, analysisName = "Pseudotime")
clusters <- colData(inSCE)$TSCAN_clusters
by.cluster <- scuttle::aggregateAcrossCells(inSCE, ids = clusters)
line.data <- TSCAN::reportEdges(by.cluster, mst = results$mst,
clusters = NULL, use.dimred = useReducedDim)
scater::plotReducedDim(inSCE, dimred = useReducedDim,
colour_by = I(colData(inSCE)$TSCAN_pseudotime),
text_by = "TSCAN_clusters", text_colour = "black") +
suppressMessages(ggplot2::scale_colour_viridis_c(name = "Pseudotime")) +
ggplot2::geom_path(data = line.data,
ggplot2::aes_string(x = names(line.data)[2],
y = names(line.data)[3],
group = 'edge'))
}
###################################################
### STEP 2:: identify expressive genes
###################################################
#' @title Test gene expression changes along a TSCAN trajectory path
#' @description Wrapper for identifying genes with significant changes with
#' respect to one of the TSCAN pseudotime paths
#' @param inSCE Input \linkS4class{SingleCellExperiment} object.
#' @param pathIndex Path index for which the pseudotime values should be used.
#' This corresponds to the terminal node of specific path from the root
#' node to the terminal node. Run \code{listTSCANTerminalNodes(inSCE)} for
#' available options.
#' @param useAssay Character. The name of the assay to use for testing the
#' expression change. Should be log-normalized. Default \code{"logcounts"}
#' @param discardCluster Cluster(s) which are not of use or masks other
#' interesting effects can be discarded. Default \code{NULL}.
#' @return The input \code{inSCE} with results updated in \code{metadata}.
#' @export
#' @author Nida Pervaiz
#' @examples
#' data("mouseBrainSubsetSCE", package = "singleCellTK")
#' mouseBrainSubsetSCE <- runTSCAN(inSCE = mouseBrainSubsetSCE,
#' useReducedDim = "PCA_logcounts")
#' terminalNodes <- listTSCANTerminalNodes(mouseBrainSubsetSCE)
#' mouseBrainSubsetSCE <- runTSCANDEG(inSCE = mouseBrainSubsetSCE,
#' pathIndex = terminalNodes[1])
runTSCANDEG <- function(inSCE,
pathIndex,
useAssay = "logcounts",
discardCluster = NULL) {
nx <- inSCE
if (!is.null(discardCluster)) {
if (any(!discardCluster %in% unique(colData(nx)$TSCAN_clusters))) {
stop("Not all `discardCluster` exist in TSCAN clusters")
}
nx <- nx[, !(colData(nx)$TSCAN_clusters %in% discardCluster)]
}
pseudo <- TSCAN::testPseudotime(nx,
pseudotime = nx[[paste0("Path_", pathIndex,
"_pseudotime")]],
assay.type = useAssay)
pseudo <- pseudo[order(pseudo$FDR),]
up.left <- pseudo[pseudo$logFC < 0,]
up.right <- pseudo[pseudo$logFC > 0,]
## Save these results in a list and then make S4 accessor that passes the
## entire list
pathresults <- list(discardClusters = discardCluster, upLeft = up.left,
upRight = up.right, useAssay = useAssay)
getTSCANResults(inSCE, analysisName = "DEG",
pathName = as.character(pathIndex)) <- pathresults
return (inSCE)
}
###################################################
### plot heatmap of top genes
###################################################
#' @title Plot heatmap of genes with expression change along TSCAN pseudotime
#' @description A wrapper function which visualizes outputs from the
#' \code{\link{runTSCANDEG}} function. Plots the top genes that change in
#' expression with increasing pseudotime along the path in the MST.
#' \code{\link{runTSCANDEG}} has to be run in advance with using the same
#' \code{pathIndex} of interest.
#' @param inSCE Input \linkS4class{SingleCellExperiment} object.
#' @param pathIndex Path index for which the pseudotime values should be used.
#' Should have being used in \code{\link{runTSCANDEG}}.
#' @param direction Should we show features with expression increasing or
#' decreeasing along the increase in TSCAN pseudotime? Choices are
#' \code{"both"}, \code{"increasing"} or \code{"decreasing"}.
#' @param topN An integer. Only to plot this number of top genes along the path
#' in the MST, in terms of FDR value. Use \code{NULL} to cancel the top N
#' subscription. Default \code{30}.
#' @param log2fcThreshold Only output DEGs with the absolute values of log2FC
#' larger than this value. Default \code{NULL}.
#' @param useAssay A single character to specify a feature expression matrix in
#' \code{assays} slot. The expression of top features from here will be
#' visualized. Default \code{NULL} use the one used for
#' \code{\link{runTSCANDEG}}.
#' @param featureDisplay Whether to display feature ID and what ID type to
#' display. Users can set default ID type by \code{\link{setSCTKDisplayRow}}.
#' \code{NULL} will display when number of features to display is less than 60.
#' \code{FALSE} for no display. Variable name in \code{rowData} to indicate ID
#' type. \code{"rownames"} or \code{TRUE} for using \code{rownames(inSCE)}.
#' @return A ComplexHeatmap in \code{.ggplot} class
#' @export
#' @author Nida Pervaiz
#' @importFrom S4Vectors metadata
#' @examples
#' data("mouseBrainSubsetSCE", package = "singleCellTK")
#' mouseBrainSubsetSCE <- runTSCAN(inSCE = mouseBrainSubsetSCE,
#' useReducedDim = "PCA_logcounts")
#' terminalNodes <- listTSCANTerminalNodes(mouseBrainSubsetSCE)
#' mouseBrainSubsetSCE <- runTSCANDEG(inSCE = mouseBrainSubsetSCE,
#' pathIndex = terminalNodes[1])
#' plotTSCANPseudotimeHeatmap(mouseBrainSubsetSCE,
#' pathIndex = terminalNodes[1])
plotTSCANPseudotimeHeatmap <- function(inSCE,
pathIndex,
direction = c("both", "increasing",
"decreasing"),
topN = 50,
log2fcThreshold = NULL,
useAssay = NULL,
featureDisplay = metadata(inSCE)$featureDisplay){
results <- getTSCANResults(inSCE, analysisName = "DEG", pathName = pathIndex)
direction <- match.arg(direction)
# Organizing cells
cell.idx <- rep(TRUE, ncol(inSCE))
if(!is.null(results$discardClusters)){
cell.idx <- cell.idx &
(!colData(inSCE)$TSCAN_clusters %in% results$discardClusters)
}
colPathPseudo <- paste0("Path_", pathIndex, "_pseudotime")
cell.idx <- cell.idx & !is.na(colData(inSCE)[[colPathPseudo]])
cell.order <- order(colData(inSCE)[[colPathPseudo]])
cell.order <- cell.order[cell.order %in% which(cell.idx)]
# Organizing genes
if (direction == "both") {
degR <- results$upRight
degL <- results$upLeft
if (!is.null(log2fcThreshold)) {
degR <- degR[abs(degR$logFC) > log2fcThreshold,]
degL <- degL[abs(degL$logFC) > log2fcThreshold,]
}
genesR <- rownames(degR)[seq(min(topN, nrow(degR)))]
genesL <- rownames(degL)[seq(min(topN, nrow(degL)))]
genesR <- genesR[!is.na(genesR)]
genesL <- genesL[!is.na(genesL)]
genes <- c(genesL, genesR)
if (length(genes) < 1) stop("No feature passed the filter.")
direction.df <- data.frame(
Direction = factor(c(rep("decreasing", length(genesL)),
rep("increasing", length(genesR))),
levels = c("increasing", "decreasing")),
row.names = genes)
} else {
if (direction == "increasing") deg <- results$upRight
if (direction == "decreasing") deg <- results$upLeft
if (!is.null(log2fcThreshold)) {
deg <- deg[abs(deg$logFC) > log2fcThreshold,]
}
topN <- min(topN, nrow(deg))
if (topN < 1) stop("No feature passed the filter.")
genes <- rownames(deg)[seq(topN)]
direction.df <- data.frame(Direction = rep(direction, length(genes)),
row.names = genes)
}
rowLabel <- featureDisplay
if (is.null(featureDisplay)) {
if (length(genes) <= 60) rowLabel <- TRUE else rowLabel <- FALSE
} else if (featureDisplay == "rownames") {
rowLabel <- TRUE
}
cellAnnotationColor = list(
.viridisPseudoTimeColor(colData(inSCE)[[colPathPseudo]])
)
names(cellAnnotationColor) <- colPathPseudo
if (is.null(useAssay)) useAssay <- results$useAssay
plotSCEHeatmap(inSCE = inSCE,
useAssay = useAssay,
cellIndex = cell.order,
featureIndex = genes,
colDend = FALSE,
rowDend = FALSE,
cluster_columns = FALSE, cluster_rows = TRUE,
colDataName = c("TSCAN_clusters", colPathPseudo),
rowLabel = rowLabel,
featureAnnotations = direction.df,
rowSplitBy = "Direction",
cellAnnotationColor = cellAnnotationColor
)
}
#' Basing on pseudotime is presented within range from 0 to 100
#' Use the viridis color scale to match with `plotTSCANResult` embedding color
#' @param x numeric vector of pseudotime
#' @return function object returned by circlize::colorRamp2
#' @noRd
.viridisPseudoTimeColor <- function(x) {
a <- max(x, na.rm = TRUE)
b <- min(x, na.rm = TRUE)
chunk <- (a - b) / 4
circlize::colorRamp2(seq(0,4) * chunk + b,
c("#440154", "#3b528b", "#21918c", "#5ec962", "#fde725"))
}
###################################################
### plot expressive genes
###################################################
#' @title Plot expression changes of top features along a TSCAN pseudotime path
#' @description A wrapper function which visualizes outputs from the
#' \code{\link{runTSCANDEG}} function. Plots the genes that increase or decrease
#' in expression with increasing pseudotime along the path in the MST.
#' \code{\link{runTSCANDEG}} has to be run in advance with using the same
#' \code{pathIndex} of interest.
#' @param inSCE Input \linkS4class{SingleCellExperiment} object.
#' @param pathIndex Path index for which the pseudotime values should be used.
#' Should have being used in \code{\link{runTSCANDEG}}.
#' @param direction Should we show features with expression increasing or
#' decreeasing along the increase in TSCAN pseudotime? Choices are
#' \code{"increasing"} or \code{"decreasing"}.
#' @param topN An integer. Only to plot this number of top genes that are
#' increasing/decreasing in expression with increasing pseudotime along
#' the path in the MST. Default 10
#' @param useAssay A single character to specify a feature expression matrix in
#' \code{assays} slot. The expression of top features from here will be
#' visualized. Default \code{NULL} use the one used for
#' \code{\link{runTSCANDEG}}.
#' @param featureDisplay Specify the feature ID type to display. Users can set
#' default value with \code{\link{setSCTKDisplayRow}}. \code{NULL} or
#' \code{"rownames"} specifies the rownames of \code{inSCE}. Other character
#' values indicates \code{rowData} variable.
#' @return A \code{.ggplot} object with the facets of the top genes. Expression
#' on y-axis, pseudotime on x-axis.
#' @export
#' @author Nida Pervaiz
#' @importFrom utils head
#' @importFrom S4Vectors metadata
#' @importFrom SummarizedExperiment rowData
#' @examples
#' data("mouseBrainSubsetSCE", package = "singleCellTK")
#' mouseBrainSubsetSCE <- runTSCAN(inSCE = mouseBrainSubsetSCE,
#' useReducedDim = "PCA_logcounts")
#' terminalNodes <- listTSCANTerminalNodes(mouseBrainSubsetSCE)
#' mouseBrainSubsetSCE <- runTSCANDEG(inSCE = mouseBrainSubsetSCE,
#' pathIndex = terminalNodes[1])
#' plotTSCANPseudotimeGenes(mouseBrainSubsetSCE,
#' pathIndex = terminalNodes[1],
#' useAssay = "logcounts")
plotTSCANPseudotimeGenes <- function(inSCE,
pathIndex,
direction = c("increasing", "decreasing"),
topN = 10,
useAssay = NULL,
featureDisplay = metadata(inSCE)$featureDisplay){
results <- getTSCANResults(inSCE, analysisName = "DEG", pathName = pathIndex)
direction = match.arg(direction)
if (!is.null(featureDisplay) &&
featureDisplay == "rownames")
featureDisplay <- NULL
if(!is.null(results$discardClusters)){
inSCE <- inSCE[,!colData(inSCE)$TSCAN_clusters %in% results$discardClusters]
}
if (direction == "decreasing") features <- rownames(results$upLeft)
if (direction == "increasing") features <- rownames(results$upRight)
features <- head(features, topN)
if (!is.null(featureDisplay)) {
features.idx <- featureIndex(features, inSCE)
features <- rowData(inSCE)[[featureDisplay]][features.idx]
}
if (is.null(useAssay)) useAssay <- results$useAssay
scater::plotExpression(inSCE,
features = features,
exprs_values = useAssay,
swap_rownames = featureDisplay,
x = paste0("Path_",pathIndex,"_pseudotime"),
colour_by = "TSCAN_clusters")
}
###################################################
### STEP3:: identify DE genes in branch cluster
###################################################
#' @title Find DE genes between all TSCAN paths rooted from given cluster
#' @description This function finds all paths that root from a given cluster
#' \code{useCluster}, and performs tests to identify significant features for
#' each path, and are not significant and/or changing in the opposite direction
#' in the other paths. Using a branching cluster (i.e. a node with degree > 2)
#' may highlight features which are responsible for the branching event. MST has
#' to be pre-calculated with \code{\link{runTSCAN}}.
#' @param inSCE Input \linkS4class{SingleCellExperiment} object.
#' @param useCluster The cluster to be regarded as the root, has to existing in
#' \code{colData(inSCE)$TSCAN_clusters}.
#' @param useAssay Character. The name of the assay to use. This assay should
#' contain log normalized counts. Default \code{"logcounts"}.
#' @param fdrThreshold Only out put DEGs with FDR value smaller than this value.
#' Default \code{0.05}.
#' @return The input \code{inSCE} with results updated in \code{metadata}.
#' @export
#' @author Nida Pervaiz
#' @examples
#' data("mouseBrainSubsetSCE", package = "singleCellTK")
#' mouseBrainSubsetSCE <- runTSCAN(inSCE = mouseBrainSubsetSCE,
#' useReducedDim = "PCA_logcounts")
#' mouseBrainSubsetSCE <- runTSCANClusterDEAnalysis(inSCE = mouseBrainSubsetSCE,
#' useCluster = 1)
runTSCANClusterDEAnalysis <- function(inSCE,
useCluster,
useAssay = "logcounts",
fdrThreshold = 0.05) {
results <- getTSCANResults(inSCE, analysisName = "Pseudotime")
message(date(), " ... Finding DEG between TSCAN branches")
tscan.pseudo <- TSCAN::orderCells(results$maptscan, mst = results$mst,
start = useCluster)
nPaths <- colnames(tscan.pseudo)
pathClusters <- list()
pathList <- data.frame()
x <- inSCE[,colData(inSCE)$TSCAN_clusters == useCluster]
store <- list()
genes <- list()
for(i in seq(nPaths)){
# Get Pseudotime of a path
pathIndex = as.character(nPaths[i])
branchPseudotime <- TrajectoryUtils::pathStat(tscan.pseudo)[,pathIndex]
colDataVarName <- paste0("TSCAN_Cluster", useCluster,
"_Path_", pathIndex, "_pseudotime")
colData(inSCE)[[colDataVarName]] <- branchPseudotime
nonNA.idx <- !is.na(branchPseudotime)
pathClusters[[pathIndex]] <- unique(colData(inSCE)$TSCAN_clusters[nonNA.idx])
pathList[i,1] <- pathIndex
pathList[i,2] <- toString(pathClusters[[pathIndex]])
message(date(), " ... Clusters involved in path index ", pathIndex,
" are: ", paste(pathClusters[[i]], collapse = ", "))
# Test along the path
pseudo.df <- TSCAN::testPseudotime(
x, df = 1,
pseudotime = branchPseudotime[colData(inSCE)$TSCAN_clusters == useCluster],
assay.type = useAssay)
pseudo.df <- pseudo.df[order(pseudo.df$p.value),]
store[[pathIndex]] <- pseudo.df
}
# Filter against all other paths
genes <- list()
for(i in seq(nPaths)){
pseudo.df <- store[[nPaths[i]]]
paths <- setdiff(seq(nPaths), i)
idx <- pseudo.df$FDR < fdrThreshold
for(j in seq_along(paths)){
pseudo.otherPath <- store[[paths[j]]]
idx <- idx & (pseudo.otherPath$p.value >= 0.05 |
sign(pseudo.otherPath$logFC) != sign(pseudo.df$logFC))
}
pseudo.df <- pseudo.df[which(idx), ]
genes[[nPaths[i]]] <- pseudo.df[order(pseudo.df$FDR),]
}
maxWidth <- max(stringr::str_length(pathList[, 1]))
choiceList <- paste0(stringr::str_pad(pathList[, 1], width=maxWidth,
side="right"),
"|", pathList[, 2])
expGenes <- list(DEgenes = genes, useAssay = useAssay,
terminalNodes = tscan.pseudo,
pathIndexList = choiceList)
getTSCANResults(inSCE, analysisName = "ClusterDEAnalysis",
pathName = useCluster) <- expGenes
return(inSCE)
}
###################################################
### plot branch cluster
###################################################
#' @title Plot TSCAN pseudotime rooted from given cluster
#' @description This function finds all paths that root from a given cluster
#' \code{useCluster}. For each path, this function plots the recomputed
#' pseudotime starting from the root on a scatter plot which contains cells only
#' in this cluster. MST has to be pre-calculated with \code{\link{runTSCAN}}.
#' @param inSCE Input \linkS4class{SingleCellExperiment} object.
#' @param useCluster The cluster to be regarded as the root, has to existing in
#' \code{colData(inSCE)$TSCAN_clusters}.
#' @param useReducedDim Saved dimension reduction name in the
#' SingleCellExperiment object. Required.
#' @param combinePlot Must be either \code{"all"} or \code{"none"}. \code{"all"}
#' will combine plots of pseudotime along each path into a single \code{.ggplot}
#' object, while \code{"none"} will output a list of plots. Default
#' \code{"all"}.
#' @return
#' \item{combinePlot = "all"}{A \code{.ggplot} object}
#' \item{combinePlot = "none"}{A list of \code{.ggplot}}
#' @export
#' @author Nida Pervaiz
#' @importFrom utils head
#' @examples
#' data("mouseBrainSubsetSCE", package = "singleCellTK")
#' mouseBrainSubsetSCE <- runTSCAN(inSCE = mouseBrainSubsetSCE,
#' useReducedDim = "PCA_logcounts")
#' plotTSCANClusterPseudo(mouseBrainSubsetSCE, useCluster = 1,
#' useReducedDim = "TSNE_logcounts")
plotTSCANClusterPseudo <- function(inSCE, useCluster, useReducedDim = "UMAP",
combinePlot = c("all", "none")){
# Get the plotting info of edges
results <- getTSCANResults(inSCE, analysisName = "Pseudotime")
combinePlot <- match.arg(combinePlot)
clusters <- colData(inSCE)$TSCAN_clusters
by.cluster <- scuttle::aggregateAcrossCells(inSCE, ids = clusters)
line.data <- TSCAN::reportEdges(by.cluster, mst = results$mst,
clusters = NULL, use.dimred = useReducedDim)
line.data.sub <- .getClustersLineData(line.data, useCluster)
# Get Branch pseudotime
tscan.pseudo <- TSCAN::orderCells(results$maptscan, mst = results$mst,
start = useCluster)
nPaths <- colnames(tscan.pseudo)
x <- inSCE[,colData(inSCE)$TSCAN_clusters == useCluster]
plotList <- list()
for (i in seq(nPaths)) {
branchPseudotime <- TrajectoryUtils::pathStat(tscan.pseudo)[,i]
branchPseudotime <- branchPseudotime[colData(inSCE)$TSCAN_clusters == useCluster]
x$pseudotime <- branchPseudotime
plotList[[i]] <- scater::plotReducedDim(x, dimred = useReducedDim,
colour_by = "pseudotime",
text_by = "TSCAN_clusters",
text_colour = "black") +
ggplot2::geom_line(data = line.data.sub,
ggplot2::aes_string(x = colnames(line.data.sub)[2],
y = colnames(line.data.sub)[3],
group = "edge"))
}
if (combinePlot == "all") {
plotList <- cowplot::plot_grid(plotlist = plotList)
}
return(plotList)
}
###################################################
### plot gene of interest in branch cluster
###################################################
#' @title Plot features identified by \code{\link{runTSCANClusterDEAnalysis}} on
#' cell 2D embedding with MST overlaid
#' @description A wrapper function which plot the top features expression
#' identified by \code{\link{runTSCANClusterDEAnalysis}} on the 2D embedding of
#' the cells cluster used in the analysis. The related MST edges are overlaid.
#' @param inSCE Input \linkS4class{SingleCellExperiment} object.
#' @param useCluster Choose a cluster used for identifying DEG with
#' \code{\link{runTSCANClusterDEAnalysis}}. Required.
#' @param pathIndex Specifies one of the branching paths from \code{useCluster}
#' and plot the top DEGs on this path. Ususally presented by the terminal
#' cluster of a path. By default \code{NULL} plot top DEGs of all paths.
#' @param useReducedDim A single character for the matrix of 2D embedding.
#' Should exist in \code{reducedDims} slot. Default \code{"UMAP"}.
#' @param topN Integer. Use top N genes identified. Default \code{9}.
#' @param useAssay A single character for the feature expression matrix. Should
#' exist in \code{assayNames(inSCE)}. Default \code{NULL} for using the one used
#' in \code{\link{runTSCANClusterDEAnalysis}}.
#' @param featureDisplay Specify the feature ID type to display. Users can set
#' default value with \code{\link{setSCTKDisplayRow}}. \code{NULL} or
#' \code{"rownames"} specifies the rownames of \code{inSCE}. Other character
#' values indicates \code{rowData} variable.
#' @param combinePlot Must be either \code{"all"} or \code{"none"}. \code{"all"}
#' will combine plots of each feature into a single \code{.ggplot} object,
#' while \code{"none"} will output a list of plots. Default \code{"all"}.
#' @return A \code{.ggplot} object of cell scatter plot, colored by the
#' expression of a gene identified by \code{\link{runTSCANClusterDEAnalysis}},
#' with the layer of trajectory.
#' @export
#' @author Yichen Wang
#' @importFrom S4Vectors metadata
#' @examples
#' data("mouseBrainSubsetSCE", package = "singleCellTK")
#' mouseBrainSubsetSCE <- runTSCAN(inSCE = mouseBrainSubsetSCE,
#' useReducedDim = "PCA_logcounts")
#' mouseBrainSubsetSCE <- runTSCANClusterDEAnalysis(inSCE = mouseBrainSubsetSCE,
#' useCluster = 1)
#' plotTSCANClusterDEG(mouseBrainSubsetSCE, useCluster = 1,
#' useReducedDim = "TSNE_logcounts")
plotTSCANClusterDEG <- function(
inSCE,
useCluster,
pathIndex = NULL,
useReducedDim = "UMAP",
topN = 9,
useAssay = NULL,
featureDisplay = metadata(inSCE)$featureDisplay,
combinePlot = c("all", "none")
) {
MSTResults <- getTSCANResults(inSCE, analysisName = "Pseudotime")
if (length(useCluster) != 1) stop("Can only specify one cluster")
DEResults <- getTSCANResults(inSCE, analysisName = "ClusterDEAnalysis",
pathName = useCluster)
if (is.null(DEResults)) {
stop("Branch DE result of specified cluster not found. Run ",
"`runTSCANClusterDEAnalysis()` with the same `useCluster` first.")
}
combinePlot <- match.arg(combinePlot)
if (is.null(pathIndex)) {
deg <- do.call(rbind, DEResults$DEgenes)
deg <- deg[order(deg$FDR),]
} else {
pathIndex <- as.character(pathIndex)
deg <- DEResults$DEgenes[[pathIndex]]
}
features <- rownames(deg)[seq(min(topN, nrow(deg), na.rm = TRUE))]
if (is.null(useAssay)) useAssay <- DEResults$useAssay
plotTSCANDimReduceFeatures(inSCE = inSCE, features = features,
useReducedDim = useReducedDim,
useAssay = useAssay, useCluster = useCluster,
featureDisplay = featureDisplay,
combinePlot = combinePlot)
}
#' @title Plot feature expression on cell 2D embedding with MST overlaid
#' @description A wrapper function which plots all cells or cells in chosen
#' cluster. Each point is a cell colored by the expression of a feature of
#' interest, the relevant edges of the MST are overlaid on top.
#' @param inSCE Input \linkS4class{SingleCellExperiment} object.
#' @param features Choose the feature of interest to explore the expression
#' level on the trajectory. Required.
#' @param useReducedDim A single character for the matrix of 2D embedding.
#' Should exist in \code{reducedDims} slot. Default \code{"UMAP"}.
#' @param useAssay A single character for the feature expression matrix. Should
#' exist in \code{assayNames(inSCE)}. Default \code{"logcounts"}.
#' @param by Where should \code{features} be found? \code{NULL},
#' \code{"rownames"} for \code{rownames(inSCE)}, otherwise will be regarded as
#' \code{rowData} variable.
#' @param useCluster Choose specific clusters where gene expression needs to be
#' visualized. By default \code{NULL}, all clusters are chosen.
#' @param featureDisplay Specify the feature ID type to display. Users can set
#' default value with \code{\link{setSCTKDisplayRow}}. \code{NULL} or
#' \code{"rownames"} specifies the rownames of \code{inSCE}. Other character
#' values indicates \code{rowData} variable.
#' @param combinePlot Must be either \code{"all"} or \code{"none"}. \code{"all"}
#' will combine plots of each feature into a single \code{.ggplot} object,
#' while \code{"none"} will output a list of plots. Default \code{"all"}.
#' @return A \code{.ggplot} object of cell scatter plot, colored by the
#' expression of a gene of interest, with the layer of trajectory.
#' @export
#' @author Yichen Wang
#' @importFrom S4Vectors metadata
#' @examples
#' data("mouseBrainSubsetSCE", package = "singleCellTK")
#' mouseBrainSubsetSCE <- runTSCAN(inSCE = mouseBrainSubsetSCE,
#' useReducedDim = "PCA_logcounts")
#' plotTSCANDimReduceFeatures(inSCE = mouseBrainSubsetSCE,
#' features = "Tshz1",
#' useReducedDim = "TSNE_logcounts")
plotTSCANDimReduceFeatures <- function(
inSCE,
features,
useReducedDim = "UMAP",
useAssay = "logcounts",
by = "rownames",
useCluster = NULL,
featureDisplay = metadata(inSCE)$featureDisplay,
combinePlot = c("all", "none"))
{
results <- getTSCANResults(inSCE, analysisName = "Pseudotime")
combinePlot <- match.arg(combinePlot)
clusters <- colData(inSCE)$TSCAN_clusters
by.cluster <- scuttle::aggregateAcrossCells(inSCE, ids = clusters)
line.data <- TSCAN::reportEdges(by.cluster, mst = results$mst,
clusters = NULL, use.dimred = useReducedDim)
if (!is.null(useCluster)) {
line.data <- .getClustersLineData(line.data, useCluster)
inSCE <- inSCE[, clusters %in% useCluster]
}
if (by == "rownames") by <- NULL
plotList <- list()
features <- stats::na.omit(features)
for (f in features) {
# Should not enter the loop if features is length zero after NA omit
g <- plotSCEDimReduceFeatures(inSCE, feature = f,
useAssay = useAssay,
featureLocation = by,dim1 = 1, dim2 = 2,
featureDisplay = featureDisplay,
reducedDimName = useReducedDim,
title = f) +
ggplot2::geom_line(data = line.data,
ggplot2::aes_string(x = colnames(line.data)[2],
y = colnames(line.data)[3],
group = "edge"),
inherit.aes = FALSE)
# Text labeling code credit: scater
reddim <- as.data.frame(SingleCellExperiment::reducedDim(inSCE,
useReducedDim))
text_out <- scater::retrieveCellInfo(inSCE, "TSCAN_clusters",
search = "colData")
by_text_x <- vapply(split(reddim[,1], text_out$val),
stats::median, FUN.VALUE = 0)
by_text_y <- vapply(split(reddim[,2], text_out$val),
stats::median, FUN.VALUE = 0)
g <- g + ggrepel::geom_text_repel(
data = data.frame(x = by_text_x,
y = by_text_y,
label = names(by_text_x)),
mapping = ggplot2::aes_string(x = "x",
y = "y",
label = "label"),
inherit.aes = FALSE, na.rm = TRUE,
)
plotList[[f]] <- g
}
if (combinePlot == "all") {
if (length(plotList) > 0)
plotList <- cowplot::plot_grid(plotlist = plotList)
}
return(plotList)
}
#' Extract edges that connects to the clusters of interest
#' @param line.data data.frame of three columns. First column should be named
#' "edge", the other two are coordinates of one of the vertices that this
#' edge connects.
#' @param useCluster Vector of clusters, that are going to be the vertices for
#' edge finding.
#' @return data.frame, subset rows of line.data
#' @noRd
.getClustersLineData <- function(line.data, useCluster) {
idx <- vapply(useCluster, function(x){
grepl(paste0("^", x, "--"), line.data$edge) |
grepl(paste0("--", x, "$"), line.data$edge)
}, logical(nrow(line.data)))
# `idx` returned was c("matrix", "array") class. Each column is a logical
# vector for edge selection of each cluster.
# Next, do "or" for each row.
idx <- rowSums(idx) > 0
line.data[idx,]
}
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.