R/clustering.R

In SydneyBioX/CiteFuse: CiteFuse: multi-modal analysis of CITE-seq data

#' spectralClustering
#'
#' A function to perform spectral clustering
#'
#' @param affinity An affinity matrix
#' @param K number of clusters
#' @param delta delta
#'
#' @return A list indicates the spectral clustering results
#'
#' @examples
#'
#' data(sce_control_subset, package = "CiteFuse")
#' sce_control_subset <- CiteFuse(sce_control_subset)
#' SNF_W <- S4Vectors::metadata(sce_control_subset)[["SNF_W"]]
#' SNF_W_clust <- spectralClustering(SNF_W, K = 5)
#'
#' @importFrom igraph arpack
#' @importFrom methods as
#'
#'
#' @export

spectralClustering <- function(affinity, 
                               K = 20, 
                               delta = 1e-5) {
    
    d <- rowSums(affinity)
    d[d == 0] <- .Machine$double.eps
    D <- diag(d)
    L <- affinity
    
    
    neff <- K + 1
    
    
    v <- sqrt(d)
    NL <- L/(v %*% t(v))
    
    
    f <- function(x, A = NULL){ # matrix multiplication for ARPACK
        as.matrix(A %*% x)
    }
    
    n <- nrow(affinity)
    
    NL <- ifelse(NL > delta, NL, 0)
    NL <- methods::as(NL, "dgCMatrix")
    
    
    eig <- igraph::arpack(f, extra = NL, sym = TRUE,
                          options = list(which = 'LA', nev = neff,
                                         n = n,
                                         ncv = max(min(c(n,4*neff)))))
    
    
    
    psi <- eig$vectors / (eig$vectors[,1] %*% matrix(1, 1, neff))#right ev
    eigenvals <- eig$values
    
    cat('Computing Spectral Clustering \n')
    
    res <- sort(abs(eigenvals), index.return = TRUE, decreasing = TRUE)
    U <- eig$vectors[, res$ix[seq_len(K)]]
    normalize <- function(x) x/sqrt(sum(x^2))
    
    U <- t(apply(U, 1, normalize))
    eigDiscrete <- .discretisation(U)
    eigDiscrete <- eigDiscrete$discrete
    labels <- apply(eigDiscrete, 1, which.max)
    
    lambda <- eigenvals[-1]/(1 - eigenvals[-1])
    lambda <- rep(1,n) %*% t(lambda)
    lam <- lambda[1,]/lambda[1,1]
    neigen <- K
    eigenvals <- eigenvals[seq_len((neigen + 1))]
    X <- psi[,2:(neigen + 1)]*lambda[, seq_len(neigen)] 
    
    return(list(labels = labels,
                eigen_values = eig$values,
                eigen_vectors = eig$vectors,
                X = X))
}




.discretisationEigenVectorData <- function(eigenVector) {
    
    Y <- matrix(0,nrow(eigenVector),ncol(eigenVector))
    maxi <- function(x) {
        i <- which(x == max(x))
        return(i[1])
    }
    j <- apply(eigenVector,1,maxi)
    Y[cbind(seq_len(nrow(eigenVector)), j)] <- 1
    
    return(Y)
    
}


.discretisation <- function(eigenVectors) {
    
    normalize <- function(x) x / sqrt(sum(x^2))
    eigenVectors <- t(apply(eigenVectors,1,normalize))
    
    n <- nrow(eigenVectors)
    k <- ncol(eigenVectors)
    
    R <- matrix(0,k,k)
    R[,1] <- t(eigenVectors[round(n/2),])
    
    mini <- function(x) {
        i <- which(x == min(x))
        return(i[1])
    }
    
    c <- matrix(0, n, 1)
    for (j in seq(2, k)) {
        c <- c + abs(eigenVectors %*% matrix(R[,j - 1], k, 1))
        i <- mini(c)
        R[,j] <- t(eigenVectors[i,])
    }
    
    lastObjectiveValue <- 0
    for (i in seq_len(1000)) {
        eigenDiscrete <- .discretisationEigenVectorData(eigenVectors %*% R)
        
        svde <- svd(t(eigenDiscrete) %*% eigenVectors)
        U <- svde[['u']]
        V <- svde[['v']]
        S <- svde[['d']]
        
        NcutValue <- 2 * (n - sum(S))
        if (abs(NcutValue - lastObjectiveValue) < .Machine$double.eps)
            break
        
        lastObjectiveValue <- NcutValue
        R <- V %*% t(U)
        
    }
    
    return(list(discrete = eigenDiscrete, continuous = eigenVectors))
}



#' reducedDimSNF
#'
#' A function to reduce the dimension of the similarity matrix
#'
#' @param sce A singlecellexperiment object
#' @param metadata indicates the meta data name of
#' affinity matrix to virsualise
#' @param method the method of visualisation, which can be UMAP,
#' tSNE and diffusion map
#' @param dimNames indicates the name of the reduced dimension results.
#'
#' @param ... other parameters for tsne(), umap()
#'
#' @return A SingleCellExperiment object
#'
#' @examples
#' data(sce_control_subset, package = "CiteFuse")
#' sce_control_subset <- CiteFuse(sce_control_subset)
#' sce_control_subset <- reducedDimSNF(sce_control_subset,
#' method = "tSNE",
#' dimNames = "tSNE_joint")
#'
#' @importFrom uwot umap
#' @importFrom Rtsne Rtsne
#' @importFrom SingleCellExperiment reducedDim
#' @importFrom S4Vectors metadata
#' @importFrom stats as.dist
#' @export

reducedDimSNF <- function(sce,
                          metadata = "SNF_W",
                          method = "UMAP",
                          dimNames = NULL,
                          ...) {
    
    method <- match.arg(method, c("UMAP", "tSNE"), several.ok = FALSE)
    
    if (!metadata %in% names(S4Vectors::metadata(sce))) {
        stop("sce does not contain metadata")
    }
    
    W <- S4Vectors::metadata(sce)[[metadata]]
    
    
    
    if (is.null(dimNames)) {
        dimNames <- paste(method, metadata, sep = "_")
    }
    
    
    if ("UMAP" %in% method) {
        
        dimred <- uwot::umap(as.dist(0.5 - W), ...)
        colnames(dimred) <- paste("UMAP", seq_len(ncol(dimred)), sep = " ")
        
        SingleCellExperiment::reducedDim(sce, dimNames) <- dimred
    }
    
    if ("tSNE" %in% method) {
        
        if (nrow(W) - 1 < 3 * 30) {
            stop("Please set a smaller perplexity number for Rtsne()")
        }
        
        dimred <- Rtsne::Rtsne(as.dist(0.5 - W), is_distance = TRUE, ...)$Y
        colnames(dimred) <- paste("tSNE", seq_len(ncol(dimred)), sep = " ")
        
        SingleCellExperiment::reducedDim(sce, dimNames) <- dimred
    }
    
    
    
    return(sce)
    
    
}






#' visualiseDim
#'
#' A function to visualise the reduced dimension
#'
#' @param sce A singlecellexperiment object
#' @param dimNames indicates the name of the reduced dimension results.
#' @param colour_by A character indicates how the cells coloured by.
#' The information either stored in colData, assay, or altExp.
#' @param shape_by A character indicates how the cells shaped by.
#' The information either stored in colData, assay, or altExp.
#' @param data_from A character indicates where the colour by data stored
#' @param assay_name A character indicates the assay name of the expression
#' @param altExp_name A character indicates the name of alternative expression
#' @param altExp_assay_name A character indicates the assay name
#' of alternative expression
#' @param dim a vector of numeric with length of 2 indicates
#' which component is being plot
#'
#' @return A ggplot of the reduced dimension visualisation
#'
#' @examples
#' data(sce_control_subset, package = "CiteFuse")
#' sce_control_subset <- CiteFuse(sce_control_subset)
#' sce_control_subset <- reducedDimSNF(sce_control_subset,
#' method = "tSNE",
#' dimNames = "tSNE_joint")
#' visualiseDim(sce_control_subset, dimNames = "tSNE_joint",
#' colour_by = "SNF_W_clust")
#'
#' @importFrom SingleCellExperiment reducedDimNames
#' @importFrom SummarizedExperiment colData
#' @importFrom S4Vectors metadata
#' @import ggplot2
#'
#' @export

visualiseDim <- function(sce,
                         dimNames = NULL,
                         colour_by = NULL,
                         shape_by = NULL,
                         data_from = c("colData", "assay", "altExp"),
                         assay_name = NULL,
                         altExp_name = NULL,
                         altExp_assay_name = NULL,
                         dim = seq_len(2)){
    
    
    if (!dimNames %in% SingleCellExperiment::reducedDimNames(sce)) {
        stop("sce does not contain dimNames")
    }
    
    data_from <- match.arg(data_from,
                           c("colData", "assay", "altExp"),
                           several.ok = TRUE)
    
    
    
    cts <- FALSE
    
    if (is.null(colour_by)) {
        
        colour_by <- NULL
        
    }else if ("character" %in% is(colour_by) & length(colour_by) == 1) {
        
        df_colour_by <- .get_color_by(sce, colour_by, data_from,
                                      assay_name, altExp_name,
                                      altExp_assay_name)
        colour_by <- df_colour_by$colour_by_info
        cts <- df_colour_by$cts
        
    } else if (length(colour_by) != ncol(sce)) {
        
        stop("colour_by needs to be a character or a vector with
               length equal to the number of cells.")
        
    } else{
        
        colour_by <- colour_by
        
    }
    
    
    if (is.null(shape_by)) {
        
        shape_by <- NULL
        
    }else if ("character" %in% is(shape_by) & length(shape_by) == 1) {
        
        if (!shape_by %in% names(colData(sce))) {
            
            stop("There is no colData with name shape_by")
            
        } else {
            
            shape_by <- as.factor(SummarizedExperiment::colData(sce)[,
                                                                     shape_by])
            
        }
        
    } else if (length(shape_by) != ncol(sce)) {
        
        stop("shape_by needs to be a character or a vector with
               length equal to the number of cells.")
        
    } else{
        
        shape_by <- shape_by
        
    }
    
    
    dimred <- reducedDim(sce, dimNames)
    
    if (is.null(colnames(dimred))) {
        colnames(dimred) <- paste(dimNames, seq_len(ncol(dimred)), sep = "_")
    }
    
    dimred <- data.frame(dimred)
    
    if (cts) {
        g_color <- scale_color_viridis_c()
    } else {
        color_values <- cite_colorPal(length(unique(colour_by)))
        g_color <- scale_color_manual(values = color_values)
    }
    
    ggplot2::ggplot(dimred, aes(x = dimred[, dim[1]],
                                y = dimred[, dim[2]],
                                col = colour_by,
                                shape = shape_by)) +
        geom_point() +
        g_color +
        scale_shape_manual(values = cite_shapePal(length(unique(shape_by)))) +
        theme_bw() +
        theme(aspect.ratio = 1) +
        xlab(colnames(dimred)[dim[1]]) +
        ylab(colnames(dimred)[dim[2]])
    
}


#' @importFrom SummarizedExperiment assay assayNames colData
#' @importFrom SingleCellExperiment altExpNames
.get_color_by <- function(sce,
                          colour_by,
                          data_from = c("colData", "assay", "altExp"),
                          assay_name = NULL,
                          altExp_name = NULL,
                          altExp_assay_name = NULL) {
    
    
    
    data_from <- match.arg(data_from,
                           c("colData", "assay", "altExp"),
                           several.ok = TRUE)
    
    colour_by_info <- NULL
    cts <- TRUE
    
    if ("colData" %in% data_from & is.null(colour_by_info)) {
        if (colour_by %in% names(colData(sce))) {
            colour_by_info <- colData(sce)[, colour_by]
            cts <- FALSE
        }
    }
    
    if ("assay" %in% data_from & is.null(colour_by_info)) {
        if (colour_by %in% rownames(sce)) {
            
            
            if (is.null(assay_name)) {
                if ("logcounts" %in% assayNames(sce)) {
                    assay_name <- "logcounts"
                } else {
                    assay_name <- assayNames(sce)[1]
                }
            } else {
                if (!assay_name %in% assayNames(sce)) {
                    stop("There is no assay_name in assayNames of the sce")
                }
            }
            colour_by_info <- assay(sce, assay_name)[colour_by, ]
            
        }
    }
    
    if ("altExp" %in% data_from &
        !is.null(altExpNames(sce))  &
        is.null(colour_by_info)) {
        if (is.null(altExp_name)) {
            if ("ADT" %in% altExpNames(sce)) {
                altExp_name <- "ADT"
            } else {
                altExp_name <- altExpNames(sce)[1]
            }
        }
        alt_se <- altExp(sce, altExp_name)
        
        if (colour_by %in% rownames(alt_se)) {
            
            if (is.null(altExp_assay_name)) {
                if ("logcounts" %in% assayNames(alt_se)) {
                    altExp_assay_name <- "logcounts"
                } else {
                    altExp_assay_name <- assayNames(alt_se)[1]
                }
            } else {
                if (!altExp_assay_name %in% assayNames(alt_se)) {
                    stop("There is no altExp_assay_name in
                           assayNames of the altExp")
                }
            }
            colour_by_info <- assay(alt_se, altExp_assay_name)[colour_by, ]
        }
    }
    
    if (is.null(colour_by_info)) {
        warning("Can not find the required colour_by info,
                        please check input data_from, assay_name,
                        altExp_name, altExp_assay_name")
    }
    return(list(colour_by_info = colour_by_info, cts = cts))
}




.normalize <- function(X) {
    row.sum.mdiag <- rowSums(X) - diag(X)
    row.sum.mdiag[row.sum.mdiag == 0] <- 1
    X <- X/(2 * (row.sum.mdiag))
    diag(X) <- 0.5
    return(X)
}

#' igraphClustering
#'
#' A function to perform igraph clustering
#'
#' @param sce A singlecellexperiment object
#' @param metadata indicates the meta data name of affinity matrix
#' to virsualise
#' @param method A character indicates the method for finding communities
#' from igraph. Default is louvain clustering.
#' @param ... Other inputs for the igraph functions
#'
#' @return A vector indicates the membership (clustering) results
#'
#' @examples
#'
#' data(sce_control_subset, package = "CiteFuse")
#' sce_control_subset <- CiteFuse(sce_control_subset)
#' SNF_W_louvain <- igraphClustering(sce_control_subset,
#' method = "louvain")
#'
#' @importFrom S4Vectors metadata
#' @importFrom dbscan sNN
#' @importFrom igraph graph_from_adjacency_matrix cluster_louvain
#' cluster_walktrap cluster_spinglass cluster_optimal
#' cluster_edge_betweenness cluster_fast_greedy cluster_label_prop
#' cluster_leading_eigen
#' @importFrom stats median as.dist
#' @export

igraphClustering <- function(sce,
                             metadata = "SNF_W",
                             method = c("louvain", "leiden", "walktrap", "spinglass",
                                        "optimal", "leading_eigen",
                                        "label_prop", "fast_greedy",
                                        "edge_betweenness"),
                             ...) {
    
    method <- match.arg(method, c("louvain", "leiden", "walktrap", "spinglass",
                                  "optimal", "leading_eigen",
                                  "label_prop", "fast_greedy",
                                  "edge_betweenness"))
    
    normalized.mat <- S4Vectors::metadata(sce)[[metadata]]
    diag(normalized.mat) <- stats::median(as.vector(normalized.mat))
    normalized.mat <- .normalize(normalized.mat)
    normalized.mat <- normalized.mat + t(normalized.mat)
    
    binary.mat <- dbscan::sNN(stats::as.dist(0.5 - normalized.mat),
                              k = 20)
    
    binary.mat <- vapply(seq_len(nrow(normalized.mat)),
                         function(x) {
                             tmp <- rep(0, ncol(normalized.mat))
                             tmp[binary.mat$id[x,]] <- 1
                             tmp
                         }, numeric(ncol(normalized.mat)))
    
    rownames(binary.mat) <- colnames(binary.mat)
    dim(binary.mat)
    
    
    g <- igraph::graph_from_adjacency_matrix(binary.mat,
                                             mode = "undirected")
    
    if (method == "louvain") {
        X <- igraph::cluster_louvain(g, ...)
    }
    
    if (method == "leiden") {
        X <- igraph::cluster_leiden(g, ...)
    }
    
    if (method == "walktrap") {
        X <- igraph::cluster_walktrap(g, ...)
    }
    
    if (method == "spinglass") {
        X <- igraph::cluster_spinglass(g, ...)
    }
    
    if (method == "optimal") {
        X <- igraph::cluster_optimal(g, ...)
    }
    
    if (method == "leading_eigen") {
        X <- igraph::cluster_leading_eigen(g, ...)
    }
    
    if (method == "label_prop") {
        X <- igraph::cluster_label_prop(g, ...)
    }
    
    if (method == "fast_greedy") {
        X <- igraph::cluster_fast_greedy(g, ...)
    }
    
    if (method == "edge_betweenness") {
        X <- igraph::cluster_edge_betweenness(g, ...)
    }
    
    clustres <- X$membership
    
    return(clustres)
    
}






#' visualiseKNN
#'
#' A function to perform louvain clustering
#'
#' @param sce A singlecellexperiment object
#' @param colour_by the name of coldata that is used to colour the node
#' @param metadata indicates the meta data name of affinity matrix to virsualise
#'
#' @return A igraph plot
#'
#' @examples
#' data(sce_control_subset, package = "CiteFuse")
#' sce_control_subset <- CiteFuse(sce_control_subset)
#' SNF_W_louvain <- igraphClustering(sce_control_subset,
#' method = "louvain")
#' visualiseKNN(sce_control_subset, colour_by = "SNF_W_louvain")
#'
#' @importFrom S4Vectors metadata
#' @importFrom dbscan sNN
#' @importFrom igraph graph_from_adjacency_matrix
#' @importFrom SummarizedExperiment colData
#' @importFrom graphics plot legend
#' @export

visualiseKNN <- function(sce,
                         colour_by = NULL,
                         metadata = "SNF_W") {
    
    
    
    normalized.mat <- S4Vectors::metadata(sce)[[metadata]]
    diag(normalized.mat) <- stats::median(as.vector(normalized.mat))
    normalized.mat <- .normalize(normalized.mat)
    normalized.mat <- normalized.mat + t(normalized.mat)
    
    binary.mat <- dbscan::sNN(as.dist(0.5 - normalized.mat), k = 20)
    
    binary.mat <- vapply(seq_len(nrow(normalized.mat)), function(x) {
        tmp <- rep(0, ncol(normalized.mat))
        tmp[binary.mat$id[x,]] <- 1
        tmp
    }, numeric(ncol(normalized.mat)))
    
    rownames(binary.mat) <- colnames(binary.mat)
    dim(binary.mat)
    
    
    g <- igraph::graph_from_adjacency_matrix(binary.mat,
                                             mode = "undirected")
    
    # cat('no. of clusters')
    # print(nlevels(as.factor(X$membership)))
    #
    if (is.null(colour_by)) {
        colour_by <- as.factor(rep(1, ncol(sce)))
    } else {
        colour_by <- as.factor(colData(sce)[, colour_by])
    }
    vertex_color <- cite_colorPal(nlevels(colour_by))[as.numeric(colour_by)]
    graphics::plot(g, vertex.label = NA,
                   edge.arrow.size = 0.000001,
                   layout = igraph::layout.fruchterman.reingold,
                   vertex.color = vertex_color,
                   vertex.size = 4)
    
    graphics::legend('bottomright',
                     legend = levels(colour_by),
                     col = cite_colorPal(nlevels(colour_by)),
                     pch = 16,
                     ncol = ceiling(nlevels(colour_by)/10),
                     cex = 0.5)
    
}