R/visualization.R

In dputhier/scigenex: The scigenex package (Single-Cell Informative GENe Explorer)

#################################################################
##    Define the plot_heatmap
#################################################################

#' @title
#' plot_heatmap
#' @description
#' Plot the observed and simulated distance with the Kth nearest neighbors.
#' @param object A ClusterSet object.
#' @param center A logical to indicate whether to center row.. 
#' @param ceil A value for ceiling (NULL for no ceiling). Ceiling is performed centering.
#' @param floor A value for flooring (NULL for no flooring). Flooring is performed after centering.
#' @param cell_clusters A vector of cell clusters with cell barcodes as names.
#' @param show_dendro A logical to indicate whether to show column dendrogram.
#' @param use_top_genes A logical to indicate whether to use highly similar genes in the slot top_genes of ClusterSet.
#' @param interactive A logical to indicate if the heatmap should be interactive.
#' @param name A title for the heatmap (if interactive is TRUE).
#' @param xlab A title for the x axis (if interactive is TRUE).
#' @param ylab A title for the y axis (if interactive is TRUE).
#' @param colorbar_name A title for the colorbar.
#' @param show_legend A logical to indicate whether to show colorbar.
#' @param colors A vector of colors.
#' @param colors_cell_clusters A named vector of colors for cell identity annotations.
#' @param row_labels A logical to indicate whether to show row labels.
#' @param col_labels A logical to indicate whether to show col labels.
#' @param label_size A value for label font size.
#' @param line_size_vertical An integer for the size of horizontal white line which separate gene clusters.
#' @param line_size_horizontal An integer for the size of vertical white line  which separate cell clusters.
#' @param link The aggloremative criterion for hierarchical clustering. One of "average", "complete" or "single". 
#' Default to average.
#' @return Iheatmap-class object.
#' @export plot_heatmap
#' @importFrom iheatmapr main_heatmap modify_layout add_row_labels add_col_labels add_col_annotation add_col_dendro add_row_title add_col_title
#' @importFrom pheatmap pheatmap
#' @examples
#' library(Seurat)
#' # Set verbosity to 1 to display info messages only.
#' set_verbosity(1)
#' 
#' # Load datasets
#' load_example_dataset('7871581/files/pbmc3k_medium_clusters')
#' load_example_dataset('7871581/files/pbmc3k_medium')
#' 
#' # Plot heatmap of 'top genes' of all gene clusters
#' pbmc3k_medium_clusters <- top_genes(pbmc3k_medium_clusters)
#' plot_heatmap(pbmc3k_medium_clusters, use_top_genes=TRUE)
#' 
#' # Plot heatmap of gene cluster 1
#' plot_heatmap(pbmc3k_medium_clusters[1,]) 
#'
#' # Plot heatmap of gene cluster 1 and 3
#' plot_heatmap(pbmc3k_medium_clusters[c(1,3),])  
#' 
#' # Plot heatmap of 'top genes' of all gene clusters
#' # with cell ordered according to Seurat results
#' plot_heatmap(pbmc3k_medium_clusters, use_top_genes=TRUE, cell_clusters=Seurat::Idents(pbmc3k_medium))
#' 
#' # Plot heatmap of 'top genes' of all gene clusters
#' # with cell ordered according to Seurat results
#' # (non interactive version)
#' plot_heatmap(pbmc3k_medium_clusters, use_top_genes=TRUE, 
#'              cell_clusters=Seurat::Idents(pbmc3k_medium),
#'              interactive=FALSE, label_size = 2)
#' @rdname plot_heatmap
plot_heatmap <- function(object,
                         center = TRUE,
                         ceil = 1,
                         floor = -1,
                         cell_clusters = NULL,
                         show_dendro = TRUE,
                         use_top_genes = FALSE,
                         interactive = TRUE,
                         name = NULL,
                         xlab = NULL,
                         ylab = NULL,
                         colorbar_name = "Exp. level",
                         show_legend = TRUE,
                         colors = colors_for_gradient("Ju1"),
                         colors_cell_clusters = NULL,
                         row_labels = TRUE,
                         col_labels = FALSE,
                         label_size = 10,
                         line_size_vertical = 3,
                         line_size_horizontal = 3,
                         link=c("average", "complete", "single", "ward.D", "ward.D2", "mcquitty")) {
  
  ## Check format object arg
  check_format_cluster_set(object)
  
  link <- match.arg(link) 
  
  check_format_cluster_set(object)
  
  if(show_dendro & is.null(cell_clusters))
    print_msg("cell_clusters is not NULL. Setting show_dendro to FALSE",
              msg_type = "INFO")
  
  if(!is.null(cell_clusters) & is.null(colors_cell_clusters))
    colors_cell_clusters <- discrete_palette(n=length(unique(cell_clusters)), "ggplot")
  
  if(!is.null(cell_clusters)){
    if(is.null(names(cell_clusters)))
      print_msg("The cell_clusters argument should be a named vector", msg_type="STOP")
    print_msg("Extracting cell identity.", msg_type="INFO")
    cell_clusters <- cell_clusters[names(cell_clusters) %in% colnames(object@data)] 
    if(inherits(cell_clusters, "factor"))
       cell_clusters <- droplevels(cell_clusters)
  }
  
  # Ensure there are enough colors
  if(length(as.character(colors_cell_clusters)) < length(table(as.character(cell_clusters))))
    print_msg("Not enough colors for cell_clusters (see colors_cell_clusters).", 
              msg_type="STOP")
  
  # Ensure there is no class equal to zero
  if(0 %in% cell_clusters){
    names_cell_clusters <- names(cell_clusters)
    cell_clusters <- as.numeric(cell_clusters)
    if(0 %in% cell_clusters){
      cell_clusters <- cell_clusters + 1
      print_msg("Found 0 in cell_clusters.", msg_type = "DEBUG")
    }
    names(cell_clusters) <- names_cell_clusters
    cell_clusters <- as.factor(cell_clusters)
  }

  print_msg(paste0("Color palette for cells: ", paste0(colors_cell_clusters, collapse=", ")),
            msg_type = "DEBUG")
  
  m <- as.matrix(object@data)
  
  # Centering
  if(center){
    print_msg("Centering matrix.", msg_type="INFO")
    m <- t(scale(t(m), center = TRUE, scale = FALSE))
  }
  
  # Ceiling and flooring
  if(!is.null(ceil)){
    print_msg("Ceiling matrix.", msg_type="INFO")
    m[m > ceil] <- ceil
  }
  
  if(!is.null(floor)){
    print_msg("Flooring matrix.", msg_type="INFO")
    m[m < floor] <- floor
  }  
  
  gene_to_clust <- gene_cluster(object)

  if(is.null(cell_clusters)){
    print_msg("Ordering cells/columns using hierarchical clustering.",
              msg_type = "INFO")
    
    # Check if sd of column is > 0
    # Otherwise HC will crash
    if(length(colnames(m)[which(apply(m, 2, sd)==0)]) > 0){
      print_msg("Some columns have sd equal to 0. Can not compute HC...", msg_type="INFO")
      print_msg("Use cell_clusters argument for ordering..", msg_type="STOP")
    }
    
    dist_cells <- cor(m, method = "pearson")
    dist_cells <- as.dist((1-dist_cells)/2)
    hclust_cells <- hclust(dist_cells, method = link)
    hclust_cells_order <- colnames(m)[hclust_cells$order]
  }
  
  if(use_top_genes) {
    print_msg("Only top genes will be used.", 
              msg_type = "INFO")
    if (length(object@top_genes) == 0) 
      print_msg("If use_top_genes is TRUE, run top_genes() before.", 
                msg_type = "STOP")
    gene_top <- unlist(object@top_genes, use.names = FALSE)
    m <- m[gene_top, ]
    gene_to_clust <- gene_to_clust[gene_top]
  }
  
  # Add blank row to separate gene clusters in the heatmap

  if(length(table(as.character(gene_to_clust))) > 1){
    
    blank_row <- matrix(NA, nrow = line_size_horizontal, ncol = ncol(m))
    
    print_msg(paste0("line_size_horizontal: ", line_size_horizontal), msg_type = "DEBUG")
    print_msg(paste0("Dim[1] blank matrix: ", nrow(blank_row)), msg_type = "DEBUG")
    print_msg(paste0("Dim[2] blank matrix: ", ncol(blank_row)), msg_type = "DEBUG")
    
    colnames(blank_row) <- colnames(m)
    m_split <- split(as.data.frame(m), gene_to_clust)
    
    nb_NA_row <- 1
    
    for(i in 1:(length(m_split)-1)){
      print_msg(paste0("Adding blank lines for cluster ", i), 
                msg_type = "DEBUG")
      rownames(blank_row) <- paste("NA.", 
                                   nb_NA_row:(nb_NA_row + line_size_horizontal - 1),
                                   sep="")
      nb_NA_row <- nb_NA_row + line_size_horizontal
      m_split[[i]] <- as.matrix(rbind(m_split[[i]], blank_row))
    }
    
    m_blank <-  do.call(rbind, m_split)
    m <- as.matrix(m_blank)
  }

  if(!is.null(cell_clusters) & length(table(as.character(cell_clusters))) > 1){
    
    # Add blank col to separate cell clusters in the heatmap
    blank_col <- matrix(NA, ncol = nrow(m), nrow = line_size_vertical)
    
    print_msg(paste0("line_size_vertical: ", line_size_vertical), msg_type = "DEBUG")
    print_msg(paste0("Dim[1] blank matrix: ", nrow(blank_col)), msg_type = "DEBUG")
    print_msg(paste0("Dim[2] blank matrix: ", ncol(blank_col)), msg_type = "DEBUG")
    
    colnames(blank_col) <- rownames(m)
    m_split <- split(as.data.frame(t(m)), cell_clusters)
    
    cell_clusters <- sort(cell_clusters)
    m <- m[ , names(cell_clusters)]
    nb_NA_row <- 1

    for(i in 1:(length(m_split)-1)){
      print_msg(paste0("Adding blank lines for cell cluster ", i), 
                msg_type = "DEBUG")
      rownames(blank_col) <- paste("NA.", 
                                   nb_NA_row:(nb_NA_row + line_size_vertical - 1),
                                   sep="")
      nb_NA_row <- nb_NA_row + line_size_vertical
      m_split[[i]] <- as.matrix(rbind(m_split[[i]], blank_col))
    }
      
      m_blank <-  do.call(rbind, m_split)
      # Now correct also the names that have been 
      # changed by do.call()...
      row.names(m_blank) <- gsub("^[^.]+.", "", rownames(m_blank), perl=T)
      m <- as.matrix(m_blank)
      # These rownames are in fact colnames (it will be transposed).
      # later on we will need the colnames to be used as rownames 
      # of the annotation_col dataframe. So we need colnames
      # to be unique. We wil add 1, 2 (...) space char... (" ").
      pos_NA <- grep("^NA\\.[0-9]+$", rownames(m), perl=T)
      row.names(m)[pos_NA] <- unlist(lapply(mapply(rep, ' ', 1:length(pos_NA)), paste0, collapse=""))
      m <- t(m)
  }
  
  # Now correct also the rownames that have been 
  # changed by do.call()
  if(length(table(as.character(gene_to_clust))) > 1){
    row.names(m) <- gsub("^[0-9]+\\.", "", rownames(m), perl=T)
    row.names(m)[grep("^NA\\.[0-9]+$", rownames(m), perl=T)] <- " "
  }

  if(is.null(cell_clusters)) {
    # This condition, if(interactive), 
    # may seem weird. In fact interactive 
    # and non interactive do not use the 
    # same plotting function. They have
    # different requirement regarding 
    # column sorting
    if(interactive)
    m <- m[, hclust_cells_order]
  }
  
  # Preparing a data.frame containing cell annotations
  if(!is.null(cell_clusters)){
    if(length(grep("^[0-9]*$", cell_clusters)) > 0){
      column <- as.factor(as.numeric(as.character(cell_clusters[colnames(m)])))
    } else {
      column <- as.factor(as.character(cell_clusters[colnames(m)]))
    }
    cell_clusters_anno <- data.frame("Ident."=column)
    rownames(cell_clusters_anno) <- colnames(m)
  }

  ####### Heatmap #######
  # Main heatmap
  print_msg("Plotting heatmap.", msg_type="INFO")
  
  if(interactive){
    
    #Flip rows
    m <- m[order(nrow(m):1),]
    
    print_msg("Plot is interactive...")
    
    htmp <- iheatmapr::main_heatmap(data = m,
                                   name = colorbar_name,
                                   show_colorbar = show_legend,
                                   colors = colors)
    
    htmp <- htmp %>% iheatmapr::modify_layout(list(margin = list(t=20, 
                                                                 r=10, 
                                                                 b=20, 
                                                                 l=20)))
    
    print_msg("Adding labels.", msg_type="DEBUG")
    
    if(row_labels){
      htmp <- htmp %>% iheatmapr::add_row_labels(font = list(size = label_size))}
    if(col_labels){
      htmp <- htmp %>% iheatmapr::add_col_labels(font = list(size = label_size))}
    
    if(!is.null(cell_clusters)){
      
      # Here, there is a bug with add_col_annotation()
      # if a single color is passed then it calls brewer.pal()
      # with the color as the palette name. To fix, we will
      # pass it 3 times the same color wich fix the bug...
      if(length(colors_cell_clusters) == 1)
        colors_cell_clusters <- rep(colors_cell_clusters, 3)

      htmp <- htmp %>% iheatmapr::add_col_annotation(cell_clusters_anno, 
                                                     colors = list("Ident." = colors_cell_clusters))
    }
    
    if(show_dendro & is.null(cell_clusters)) {
      htmp <- htmp %>% iheatmapr::add_col_dendro(hclust_cells, reorder = FALSE)
    }

    print_msg("Adding Titles.", msg_type="DEBUG")
    
    if(!is.null(ylab)){
      htmp <- htmp %>% add_row_title(ylab, side="right", font = list(size = 12))
      }
    if(!is.null(xlab)){
      htmp <- htmp %>% add_col_title(xlab, side="top", font = list(size = 12))
      }
    if(!is.null(name)){
      htmp <- htmp %>% add_col_title(name, side="top", font = list(size = 24))
      }
  }else{
    # Reorder rows to get the same order as the interactive heatmap
    #m <- m[order(nrow(m):1),]
    
    cluster_cols <- F
    annotation_col <- NA
    
    if(is.null(cell_clusters)){
      if(show_dendro){
        cluster_cols <- hclust_cells
      }
    }

    if(!is.null(cell_clusters)){
      annotation_col <- cell_clusters_anno
    }

    htmp <- pheatmap::pheatmap(mat = m, 
                     annotation_legend = show_legend, legend = show_legend,
                     color = colorRampPalette(colors)(100),
                     cluster_rows = FALSE, 
                     cluster_cols = cluster_cols, 
                     fontsize_row= label_size,
                     show_rownames = row_labels, 
                     show_colnames = col_labels, 
                     annotation_col = annotation_col,
                     border_color = NA, 
                     scale = "none", 
                     annotation_colors=list("Ident."=setNames(colors_cell_clusters, 
                                                                    unique(as.character(sort(cell_clusters))))),
                     na_col = "white")
    
    if(!is.null(ylab)){
      htmp <- htmp %>% add_row_title(ylab, side="right", font = list(size = 12))
    }
    if(!is.null(xlab)){
      htmp <- htmp %>% add_col_title(xlab, side="top", font = list(size = 12))
    }
    if(!is.null(name)){
      htmp <- htmp %>% add_col_title(name, side="top", font = list(size = 24))
    }
    
  }
  

  return(htmp)
}

#################################################################
##    Define the plot_dist function
#################################################################

#' Plot Distribution of distances
#' 
#' This function creates a histogram of the simulated and observed distances 
#' of a ClusterSet object.
#' 
#' @param object A ClusterSet object.
#' @param bins The number of bins to use in the histogram (default is 150).
#' @param alpha The level of transparency for the bars in the histogram
#'  (default is 0.5).
#' @param colors A named vector of colors to use for the histograms, 
#' with "Simulated" and "Observed" as the named elements 
#' (default is c("Simulated" = "#FB8500", "Observed" = "#36949D")).
#' @param xlim Limits for the x axis of the histogram (e.g. c(0.8, 1)).
#' @param vline_color Color of the vertical line indicating 
#' the critical distance with KNN.
#' @param text_size Font size for the label of the critical distance.
#' @param text_hjust Horizontal justification for the label 
#' of the critical distance.
#' @param text_vjust Vertical justification for the label 
#' of the critical distance.
#' 
#' @return A ggplot object.
#' 
#' @examples
#' # Set verbosity to 1 to display info messages only.
#' set_verbosity(1)
#' 
#' # Load datasets
#' load_example_dataset('7871581/files/pbmc3k_medium_clusters')
#' plot_dist(pbmc3k_medium_clusters)
#' 
#' @export plot_dist
#' 
plot_dist <- function(object,
                      bins=150,
                      alpha=0.5,
                      colors=c("Simulated"="#FB8500", "Observed"="#36949D"),
                      xlim=NULL,
                      vline_color = "#4f6d7a",
                      text_size = 4,
                      text_hjust = -0.8,
                      text_vjust = -0.5) {
  
  ## Check format object arg
  check_format_cluster_set(object)
  
  DKNN = c(object@dbf_output$simulated_dknn,
           object@dbf_output$dknn)
  Type = c(rep("Simulated",
               length(object@dbf_output$simulated_dknn)),
           rep("Observed",
               length(object@dbf_output$dknn)))
  df <- data.frame(DKNN, Type)
  p <-  ggplot(df, aes(x=DKNN, fill=Type)) +
    geom_histogram(bins=bins, 
                   position="identity", 
                   alpha=alpha, 
                   color="white") + 
    theme_bw() + 
    scale_fill_manual(values=colors) +
    theme(panel.grid.minor = element_blank(),
          panel.grid.major.x = element_blank(),
          axis.line = element_line(colour = "black"),
          panel.border = element_blank(),
          axis.title = element_text(size = 15),
          axis.text = element_text(size = 10)) +
    xlab(label = "Distance with KNN") +
    ylab(label = "Count")  +
    geom_vline(aes(xintercept = object@dbf_output$critical_distance),
               color = vline_color,
               linetype = "dotdash") +
    geom_text(mapping = aes(x = object@dbf_output$critical_distance,
                            y = 0,
                            label = "Critical distance with KNN",
                            hjust = text_hjust,
                            vjust = text_vjust,
                            angle = 90),
              color = vline_color,
              size = text_size)
  
  if(!is.null(xlim))
    p <- p + xlim(xlim)
  
  return(p)
}


# -------------------------------------------------------------------------
# plot_ggheatmap ---------------------------------------------------------
# -------------------------------------------------------------------------

#' @title
#' Heatmap of data contained in a ClusterSet object.
#' @description
#' Plot the results (heatmap) contained in a ClusterSet object.
#' @param object A ClusterSet object.
#' @param to_log2 Whether data should be transform in logarithm base 2 (+ 1 as a pseudocount).
#' @param use_top_genes A logical to indicate whether to use highly similar genes in the slot top_genes of ClusterSet.
#' @param ident A named vector containing the cell type identities for each cell.
#' Typically the result from the Idents() function on a Seurat object (see Seurat library). Lists are
#' also accepted which will results in multi-level facets.
#' @param panel_spacing Spacing between facets/panels ("line" units).
#' @param colors A vector of colors for the gradient.
#' @param standardizing Whether rows should be divided by standard deviation.
#' @param color_ident A vector containing colors for the cell classes as. 
#' also accepted which will results in multi-level facets.
#' @param ceil A value for ceiling (NULL for no ceiling). Ceiling is performed after log transformation, centering and standardization.
#' @param floor A value for flooring (NULL for no flooring). Flooring is performed after log transformation, centering and standardization.
#' @param centering Whether rows should be centered. 
#' @param xlab A name for the x axis.
#' @param ylab A name for the y axis.
#' @param hide_gene_name Whether to hide gene names.
#' @param hide_col_name Whether to hide column names.
#' 
#' @return A ggplot diagram.
#' @export plot_ggheatmap
#'
#' @examples
#' library(Seurat)
#' # Load datasets
#' load_example_dataset('7871581/files/pbmc3k_medium_clusters')
#' load_example_dataset('7871581/files/pbmc3k_medium')
#' 
#' # rename clusters
#' new_obj <- rename_clust(pbmc3k_medium_clusters, new_name=sprintf("M%02d", as.integer(clust_names(pbmc3k_medium_clusters))))
#' 
#' # Use plot_ggheatmap
#' ident_pbmc3k <- sort(Seurat::Idents(pbmc3k_medium))
#' new_obj <- top_genes(new_obj)
#' plot_ggheatmap(new_obj[,names(ident_pbmc3k)], ident=ident_pbmc3k)
#' @keywords internal
setGeneric("plot_ggheatmap",
           
           function(object,
                    to_log2 = FALSE,
                    use_top_genes=TRUE,
                    ident=NULL,
                    panel_spacing=0.05,
                    colors = colors_for_gradient("Ju1"),
                    standardizing = FALSE,
                    color_ident=NULL,
                    ceil=1,
                    floor=-1,
                    centering = TRUE,
                    xlab="Genes",
                    ylab="Spots",
                    hide_gene_name=TRUE,
                    hide_col_name=TRUE,
                    pseudocount=0.1) {
             
             standardGeneric("plot_ggheatmap")
           })

#' @title
#' Heatmap of data contained in a ClusterSet object.
#' @description
#' Plot the results (heatmap) contained in a ClusterSet object.
#' @param object A ClusterSet object.
#' @param to_log2 Whether data should be transform in logarithm base 2 (+ 0.1 as a pseudocount).
#' @param use_top_genes A logical to indicate whether to use highly similar genes in the slot top_genes of ClusterSet.
#' @param ident A named vector containing the cell type identities for each cell.
#' Typically the result from the Idents() function on a Seurat object (see Seurat library).
#' @param panel_spacing Spacing between facets/panels ("line" units).
#' @param colors A vector of colors for the gradient.
#' @param standardizing Whether rows should be divided by standard deviation.
#' @param color_ident A vector containing colors for the cell classes as. 
#' @param ceil A value for ceiling (NULL for no ceiling). Ceiling is performed after log transformation, centering and standardization.
#' @param floor A value for flooring (NULL for no flooring). Flooring is performed after log transformation, centering and standardization.
#' @param centering Whether rows should be centered. 
#' @param xlab A name for the x axis.
#' @param ylab A name for the y axis.
#' @param hide_gene_name Whether to hide gene names.
#' @param hide_col_name Whether to hide column names.
#' @param pseudocount A value for the pseudocount added before log transformation.
#' @return A ggplot diagram.
#' @export plot_ggheatmap
#' @importFrom reshape2 melt
#' @importFrom ggh4x facet_grid2 strip_themed elem_list_rect
#' @importFrom ggplot2 ggplot geom_raster theme_bw scale_fill_gradientn theme facet_grid element_blank element_rect element_text
#' @examples
#' library(Seurat)
#' # Load datasets
#' load_example_dataset('7871581/files/pbmc3k_medium_clusters')
#' load_example_dataset('7871581/files/pbmc3k_medium')
#' 
#' # rename clusters
#' new_obj <- rename_clust(pbmc3k_medium_clusters, new_name=sprintf("M%02d", as.integer(clust_names(pbmc3k_medium_clusters))))
#' 
#' # Use plot_ggheatmap
#' ident_pbmc3k <- sort(Seurat::Idents(pbmc3k_medium))
#' new_obj <- top_genes(new_obj)
#' plot_ggheatmap(new_obj[,names(ident_pbmc3k)], ident=ident_pbmc3k)
setMethod(
  "plot_ggheatmap",
  signature(object = "ClusterSet"),
  function(object,
           to_log2 = FALSE,
           use_top_genes=TRUE,
           ident=NULL,
           panel_spacing=0.05,
           colors = colors_for_gradient("Ju1"),
           standardizing = FALSE,
           color_ident=NULL,
           ceil=1,
           floor=-1,
           centering = TRUE,
           xlab="Spots",
           ylab="Genes",
           hide_gene_name=TRUE,
           hide_col_name=TRUE,
           pseudocount=0.1) {
    
    check_format_cluster_set(object)
    print_msg("getting matrix", msg_type="DEBUG")
    
    if(!is.list(ident)){
      
      name_idents <- names(ident)
      
      if(is.null(name_idents)){
        print_msg("The 'ident' argument needs a named vector or a named list of named vector.")
      }
      
      if(length(which(names(ident) == "")) != 0){
        print_msg("The 'ident' argument needs a named vector or a named list of named vector.")
      }
      
    }
    
    nb <- nclust(object)
    
    if(use_top_genes){
      if(length(object@top_genes) == 0)
        print_msg("Please use top_gene() methods onto ClusterSet object.", msg_type = "STOP")
      m <- object@data[unlist(object@top_genes), ]
    }else{
      m <- object@data
    }
    
    
    if(to_log2) {
      m <- log2(m + pseudocount)
    }
    
    ## median-centering of row
    if (centering) {
      print_msg("Median-centering rows.", msg_type="DEBUG")
      mean_row <- apply(m, 1, mean)
      m <- sweep(m, MARGIN = 1, STATS = mean_row, FUN = "-")
    }
    
    ## Standardizing row
    if (standardizing) {
      print_msg("Standardizing rows.", msg_type="DEBUG")
      sd_row <- apply(m, 1, sd)
      m <- sweep(m, MARGIN = 1, STATS = sd_row, FUN = "/")
    }
    
    ## Ceiling / flooring
    if(!is.null(ceil)){
      print_msg("Ceiling matrix.", msg_type="DEBUG")
      m[m > ceil] <- ceil
    }
    
    if(!is.null(floor)){
      print_msg("Flooring matrix.", msg_type="DEBUG")
      m[m < floor] <- floor
    }
    
    ## melting
    print_msg("Melting matrix.", msg_type="DEBUG")
    m_melt <- reshape2::melt(as.matrix(m))
    colnames(m_melt) <- c("genes", "samples", "values")
    gclust <- gene_cluster(object, as_string = T)
    match_gclust <- match(m_melt$genes, names(gclust))
    
    lev_clust <- suppressWarnings(sort(unique(as.numeric(gclust))))

    if(length(lev_clust) == 0){
      lev_clust <- unique(gclust)
      m_melt$gene_clusters <- factor(gclust[match_gclust], levels = lev_clust, ordered = TRUE)
    }else{
      m_melt$gene_clusters <- factor(gclust[match_gclust], levels = lev_clust, ordered = TRUE)
    }
    
    

    if(!is.null(ident)){
      
      if(!is.list(ident)){
        
        if(!all(m_melt$cell %in% names(ident))){
          print_msg("All cell need a target cluster when using 'ident'.", msg_type = "STOP")
        }
        m_melt$cell_clusters <- factor(ident[m_melt$samples], ordered = TRUE)
      
      }else{
        
        for(i in 1:length(names(ident))){
          if(!all(m_melt$cell %in% names(ident[[i]]))){
            print_msg("All cell need a target cluster when using 'ident'.", msg_type = "STOP")
          }
          m_melt[, names(ident[i])] <- factor(ident[[i]][m_melt$samples], ordered = TRUE)
        }
        
    }
      
    }
    
    
    ## plotting
    # Note that samples, value, gene, cluster
    # may appear as undefined variable to "R check" command.
    # A workaround is to define them as NULL first...
    samples <- values <- genes <- cell_clusters <- gene_clusters <- NULL

    print_msg("Preparing diagram (tile).", msg_type="DEBUG")
    print_msg(paste0("Matrix columns :", colnames(m_melt)), msg_type="DEBUG")

    p <- ggplot2::ggplot(
      data = m_melt,
      aes(
        x = samples,
        y = genes,
        fill = values
      )
    )

    print_msg("Preparing color palette.", msg_type="DEBUG")
    color.ramp <- grDevices::colorRampPalette(colors)(10)
    
    p <- p + ggplot2::geom_raster()
    p <- p + ggplot2::theme_bw()
    p <- p + ggplot2::scale_fill_gradientn(
      colours = color.ramp,
      name = "Signal"
    )
    
    nb_cell_classes <- length(table(m_melt$cell_clusters))

    p <- p + ggplot2::xlab(xlab) + ggplot2::ylab(ylab)
    
    print_msg("Theming.", msg_type="DEBUG")
    p <- p + ggplot2::theme(
      axis.ticks.y = ggplot2::element_blank(),
      axis.ticks.x = ggplot2::element_blank(),
      panel.spacing = grid::unit(0.05, "lines"),
      panel.border = ggplot2::element_blank(),
      strip.background.x = ggplot2::element_rect(colour = "white"),
      strip.background.y = ggplot2::element_rect(fill = "#444444", colour = "white"),
      strip.text.y = ggplot2::element_text(colour = "white", angle = 0),
    )
    
    if(hide_gene_name){
      p <- p + ggplot2::theme(axis.text.y = ggplot2::element_blank())
    }
    
    if(hide_col_name){
      p <- p + ggplot2::theme(axis.text.x = ggplot2::element_blank())
    }
    print_msg("Adding facets.", msg_type="DEBUG")
    
    gg_color_hue <- function(n) {
      hues = seq(15, 375, length = n + 1)
      hcl(h = hues, l = 65, c = 100)[1:n]
    }
    
    
    if(!is.null(ident)){
      if(!is.list(ident)){
        if(is.null(color_ident)){
          color_ident <- gg_color_hue(nb_cell_classes)
        }else{
          if(length(color_ident) != nb_cell_classes){
            print_msg("Need as many colors as cell classes", msg_type = "STOP")
          }
        }
          
        colored_strip <- ggh4x::strip_themed(background_x = ggh4x::elem_list_rect(fill = color_ident))
        p <- p + ggh4x::facet_grid2(gene_clusters ~ cell_clusters, 
                                    scales = "free", space = "free",
                                    strip=colored_strip)
      }else{
        p <- p + ggplot2::facet_grid(as.formula(paste0("gene_clusters ~ ",  paste0(names(ident), collapse = " + "))), 
                                    scales = "free", space = "free")
      }

                                  
    }else{
      p <- p + ggplot2::facet_grid(gene_clusters ~ ., scales = "free", space = "free")
    }
    
    return(p)
})