R/scScoreDimPlot.R

Defines functions scScoreDimPlot

Documented in scScoreDimPlot

#' MSigdb scoring DimPlot for single-cell clusters
#' @title scScoreDimPlot
#' @param object Seurat object
#' @param signature Name of one gene set
#' @param reduction Which dimensionality reduction to use
#' @param cols Colors to plot
#' @param pt.size Adjust point size to plot, default pt.size=0.5
#' @param split.by Name of a metadata column to split plot by
#' @param title Title of the plot
#' @param ncol Number of columns for display the plots
#' @param raster Convert points to raster format, default is TRUE
#' @param scale Determine whether to scale the data, default is TRUE
#' @param col.min Minimum scaled average score threshold 
#'   (smaller values will be set to this)
#' @param col.max Maximum scaled average score threshold 
#'   (larger values will be set to this)
#' @return ggplot2 object
#' @export

scScoreDimPlot <- function(object = NULL, signature = NULL, reduction = NULL, 
                           cols = NULL, pt.size = NULL, split.by = NULL, 
                           title = NULL, ncol = NULL, raster = TRUE, 
                           scale = TRUE, col.min = NA, col.max = NA) {
  if (is.null(signature)) {
    stop("Parameter 'signature' must be specified!\n")
  } else if (length(signature) != 1) {
    stop("Parameter 'signature' must be one pathway!\n")
  }
  if (is.null(reduction)) {
    if ("umap" %in% names(object)) {
      reduction <- "umap"
    } else if ("tsne" %in% names(object)) {
      reduction <- "tsne"
    } else {
      stop("The reduction parameter does not support! 
        Please use 'umap', 'tsne', or 'pca' instead.\n")
    }
  }
  if (is.null(cols)) {
      cols <- colorRampPalette(rev(RColorBrewer::brewer.pal(11, "Spectral")))(32)
  }
  if (is.null(pt.size)) {
    pt.size <- 0.5
  }
  xmin <- xmax <- ymin <- ymax <- NULL
  ps <- function(data, min.value = min.value, max.value = max.value, 
                 title = NULL, legend_title = NULL) {
    data <- suppressWarnings(data.frame(apply(data, 2, as.numeric)))
    p <- ggplot2::ggplot(data = data, ggplot2::aes(x = data[, 1], 
      y = data[, 2], z = data[, 3]))
    if (isTRUE(raster)) {
      p <- p + ggrastr::rasterise(dpi = 300, 
        ggplot2::geom_point(ggplot2::aes(colour = data[, 3]), 
          size = pt.size, na.rm = TRUE))
    } else {
      p <- p + ggplot2::geom_point(ggplot2::aes(colour = data[, 3]), 
        size = pt.size, na.rm = TRUE)
    }
    p <- p + ggplot2::scale_colour_gradientn(colours = cols, na.value = "white",
      breaks = seq(min.value, max.value, length = 5), limit = c(min.value,
        max.value), guide = ggplot2::guide_colorbar(title = legend_title,
        order = 1, title.theme = ggplot2::element_text(size = 8), 
        label.theme = ggplot2::element_text(size = 8)))
    p <- p + cowplot::theme_cowplot() + 
      ggplot2::labs(x = colnames(data)[1], y = colnames(data)[2], title = title)
    p <- p + ggplot2::scale_x_continuous(limits = c(xmin, xmax), 
      breaks = seq(floor(xmin/5) * 5, ceiling(xmax/5) * 5, by = 5))
    p <- p + ggplot2::scale_y_continuous(limits = c(ymin,
      ymax), breaks = seq(floor(ymin/5) * 5, ceiling(ymax/5) * 5, by = 5))
    return(p)
  }
  pm <- function(data, min.value = min.value, max.value = max.value, 
                 title = NULL, legend_title = NULL) {
    data <- suppressWarnings(data.frame(apply(data, 2, as.numeric)))
    p <- ggplot2::ggplot(data = data, ggplot2::aes(x = data[, 1], 
      y = data[, 2], z = data[, 3]))
    if (isTRUE(raster)) {
      p <- p + ggrastr::rasterise(dpi = 300, 
        ggplot2::geom_point(ggplot2::aes(colour = data[, 3]), 
          size = pt.size, na.rm = TRUE))
    } else {
      p <- p + ggplot2::geom_point(ggplot2::aes(colour = data[, 3]), 
        size = pt.size, na.rm = TRUE)
    }
    p <- p + ggplot2::scale_colour_gradientn(colours = cols, 
      na.value = "white", breaks = seq(min.value, max.value, length = 5), 
      limit = c(min.value, max.value))
    p <- p + cowplot::theme_cowplot() + ggplot2::labs(x = colnames(data)[1],
      y = colnames(data)[2], title = title)
    p <- p + ggplot2::scale_x_continuous(limits = c(xmin, xmax), 
      breaks = seq(floor(xmin/5) * 5, ceiling(xmax/5) * 5, by = 5))
    p <- p + ggplot2::scale_y_continuous(limits = c(ymin, ymax), 
      breaks = seq(floor(ymin/5) * 5, ceiling(ymax/5) * 5, by = 5))
    p <- p + ggplot2::theme(legend.position = "none")
    return(p)
  }
  reduction_ids <- gsub("coord", toupper(reduction), c("coord_1", "coord_2"))
  if (is.null(x = split.by)) {
    Data <- object@meta.data[, c(reduction_ids, signature)]
    Data[, signature][is.na(Data[, signature])] <- 0
    if (is.na(col.min)) {
      col.min <- round(min(Data[, signature]), 1)
    }
    if (is.na(col.max)) {
      col.max <- round(max(Data[, signature]), 1)
    }
    Data[, signature][Data[, signature] > col.max] <- col.max
    Data[, signature][Data[, signature] < col.min] <- col.min
    xmin <- min(Data[, 1])
    xmax <- max(Data[, 1])
    ymin <- min(Data[, 2])
    ymax <- max(Data[, 2])
    return(ps(data = Data, min.value = col.min, max.value = col.max, 
      title = title, legend_title = signature))
  }
  plots <- list()
  if (!is.null(x = split.by)) {
    if (!split.by %in% colnames(object@meta.data)) {
      stop("The parameter 'split.by' ", split.by, " does not exist in MetaData slot!\n")
    }
    Data <- object@meta.data[, c(reduction_ids, signature, split.by)]
    if (scale) {
      Data[, signature] <- scale(x = Data[, signature])
    }
    Data[, signature][is.na(Data[, signature])] <- 0
    if (is.na(col.min)) {
      col.min <- round(min(Data[, signature]), 1)
    }
    if (is.na(col.max)) {
      col.max <- round(max(Data[, signature]), 1)
    }
    Data[, signature][Data[, signature] > col.max] <- col.max
    Data[, signature][Data[, signature] < col.min] <- col.min
    xmin <- min(Data[, 1])
    xmax <- max(Data[, 1])
    ymin <- min(Data[, 2])
    ymax <- max(Data[, 2])
    if (is.null(ncol)) {
      ncol = ceiling(sqrt(length(unique(Data[, split.by]))))
    }
    legend <- ps(data = Data, min.value = col.min, max.value = col.max, 
      title = NULL, legend_title = signature)
    legend <- ggplot2::ggplotGrob(legend)
    legend <- gtable::gtable_filter(legend, "box", trim = FALSE)
    for (s in unique(Data[, split.by])) {
      data <- Data[Data[, split.by] == s, ]
      plots[[s]] <- pm(data, min.value = col.min, max.value = col.max, title = s,
        legend_title = NULL)
        }
    }
    return((patchwork::wrap_plots(plots, ncol = ncol) | legend) + 
            patchwork::plot_layout(widths = c(3, 1)))
}
ncrna/Yeskit.old documentation built on Dec. 22, 2021, 12:06 a.m.