stacked barplot for replicate metrics

[kelvinyevng]

I have seen stacked barplots in several papers presenting single cell data. Representation of replicate information on a per cluster basis seems to be advantageously presented in this fashion. Is it possible to perform in conos?

Like this:

[evanbiederstedt]

Sounds like a custom analysis you should do. https://ggplot2.tidyverse.org/ will be useful.

[kelvinyevng]

There is a function in the conos package that is called:
plotClusterBarplots(con, legend.height = 0.1)
but it shows for each sample, is it possible to make it so it shows for each condition instead?
If you could have a condition called: BM and one called CB instead of BM1, BM2, CB1, CB2.

[evanbiederstedt]

Yes, it's very possible.

conos/R/plot.R

Lines 148 to 224 in e9f4b04

	#' Plots barplots per sample of composition of each pagoda2 application based on
	#' selected clustering
	#'
	#' @param conos.obj A conos object (default=NULL)
	#' @param clustering name of clustering in the current object (default=NULL)
	#' @param groups arbitrary grouping of cells (to use instead of the clustering) (default=NULL)
	#' @param sample.factor a factor describing cell membership in the samples (or some other category) (default=NULL). This will default to samples if not provided.
	#' @param show.entropy boolean Whether to include entropy barplot (default=TRUE)
	#' @param show.size boolean Whether to include size barplot (default=TRUE)
	#' @param show.composition boolean Whether to include composition barplot (default=TRUE)
	#' @param legend.height numeric Relative hight of the legend panel (default=0.2)
	#' @return a ggplot object
	#' @export
	plotClusterBarplots <- function(conos.obj=NULL, clustering=NULL, groups=NULL, sample.factor=NULL, show.entropy=TRUE, show.size=TRUE, show.composition=TRUE, legend.height=0.2) {
	## param checking
	if (!is.null(clustering) && is.null(conos.obj)) {
	stop('conos.obj must be passed if clustering name is specified')
	}
	if (is.null(groups) && is.null(conos.obj)) {
	stop('Either groups factor on the cells or a conos object needs to be specified, both cannot be NULL')
	}
	groups <- parseCellGroups(conos.obj, clustering, groups)
	if (!is.null(groups)) {
	groups <- as.factor(groups)
	}
	if(is.null(sample.factor)) {
	sample.factor <- conos.obj$getDatasetPerCell() # assignment to samples
	}
	xt <- table(sample.factor[match(names(groups),names(sample.factor))],groups)
	xt <- xt[rowSums(xt)>0, ]
	xt <- xt[ ,colSums(xt)>0]
	df <- reshape2::melt(xt)
	colnames(df) <- c("sample","cluster","f")
	df$f <- df$f/colSums(xt)[as.character(df$cluster)]
	clp <- ggplot2::ggplot(df, ggplot2::aes(x=factor(.data$cluster, levels=levels(groups)),y=.data$f,fill=.data$sample)) +
	ggplot2::geom_bar(stat='identity') + ggplot2::xlab('cluster') + ggplot2::ylab('fraction of cells') + ggplot2::theme_bw() +
	ggplot2::scale_y_continuous(expand=c(0, 0))
	if (!show.size && !show.entropy){
	return(clp)
	}
	# extract legend
	leg <- cowplot::get_legend(clp + ggplot2::theme(legend.position="bottom"))
	pl <- list(clp + ggplot2::theme(legend.position="none"))
	if(show.entropy) {
	if (!requireNamespace("entropy", quietly=TRUE)){
	stop("You need to install 'entropy' package to use 'show.entropy=TRUE'")
	}
	n.samples <- nrow(xt)
	ne <- 1-apply(xt, 2, entropy::KL.empirical, y2=rowSums(xt), unit=c('log2')) / log2(n.samples) # relative entropy
	enp <- ggplot2::ggplot(data.frame(cluster=factor(colnames(xt),levels=levels(groups)),entropy=ne), ggplot2::aes(.data$cluster, .data$entropy)) +
	ggplot2::geom_bar(stat='identity',fill='grey65') + ggplot2::ylim(0,1) +
	ggplot2::geom_hline(yintercept=1, linetype="dashed", color = "grey30") + ggplot2::theme_bw()
	pl <- c(pl,list(enp))
	}
	if(show.size) {
	szp <- ggplot2::ggplot(data.frame(cluster=factor(colnames(xt),levels=levels(groups)), cells=colSums(xt)), ggplot2::aes(.data$cluster, .data$cells)) +
	ggplot2::geom_bar(stat='identity') + ggplot2::scale_y_continuous(trans='log10') + ggplot2::theme_bw() + ggplot2::ylab('number of cells')
	pl <- c(pl, list(szp))
	}
	pp <- cowplot::plot_grid(plotlist=pl ,ncol=1, rel_heights=c(1,rep(0.3,length(pl)-1)))
	pp2 <- cowplot::plot_grid(leg, pp, ncol=1, rel_heights=c(legend.height,1))
	return(pp2)
	}

plotClusterBarplots <- function(conos.obj=NULL, clustering=NULL, groups=NULL, sample.factor=NULL, show.entropy=TRUE, show.size=TRUE, show.composition=TRUE, legend.height=0.2) {
  ## param checking
  if (!is.null(clustering) && is.null(conos.obj)) {
    stop('conos.obj must be passed if clustering name is specified')
  }

  if (is.null(groups) && is.null(conos.obj)) {
    stop('Either groups factor on the cells or a conos object needs to be specified, both cannot be NULL')
  }

  groups <- parseCellGroups(conos.obj, clustering, groups)

  if (!is.null(groups)) {
    groups <- as.factor(groups)
  }

  if(is.null(sample.factor)) {
    sample.factor <- conos.obj$getDatasetPerCell() # assignment to samples
  }

  xt <- table(sample.factor[match(names(groups),names(sample.factor))],groups)
  xt <- xt[rowSums(xt)>0, ]
  xt <- xt[ ,colSums(xt)>0]

  df <- reshape2::melt(xt)
  colnames(df) <- c("sample","cluster","f")
  df$f <- df$f/colSums(xt)[as.character(df$cluster)]
  clp <- ggplot2::ggplot(df, ggplot2::aes(x=factor(.data$cluster, levels=levels(groups)),y=.data$f,fill=.data$sample)) +
    ggplot2::geom_bar(stat='identity') + ggplot2::xlab('cluster') + ggplot2::ylab('fraction of cells') + ggplot2::theme_bw() +
    ggplot2::scale_y_continuous(expand=c(0, 0))

  if (!show.size && !show.entropy){
    return(clp)
  }

  # extract legend
  leg <- cowplot::get_legend(clp + ggplot2::theme(legend.position="bottom"))
  pl <- list(clp + ggplot2::theme(legend.position="none"))

  if(show.entropy) {
    if (!requireNamespace("entropy", quietly=TRUE)){
      stop("You need to install 'entropy' package to use 'show.entropy=TRUE'")
    }

    n.samples <- nrow(xt)
    ne <- 1-apply(xt, 2, entropy::KL.empirical, y2=rowSums(xt), unit=c('log2')) / log2(n.samples) # relative entropy
    enp <- ggplot2::ggplot(data.frame(cluster=factor(colnames(xt),levels=levels(groups)),entropy=ne), ggplot2::aes(.data$cluster, .data$entropy)) +
      ggplot2::geom_bar(stat='identity',fill='grey65') + ggplot2::ylim(0,1) +
      ggplot2::geom_hline(yintercept=1, linetype="dashed", color = "grey30") + ggplot2::theme_bw()
    pl <- c(pl,list(enp))
  }

  if(show.size) {
    szp <- ggplot2::ggplot(data.frame(cluster=factor(colnames(xt),levels=levels(groups)), cells=colSums(xt)), ggplot2::aes(.data$cluster, .data$cells)) +
      ggplot2::geom_bar(stat='identity') + ggplot2::scale_y_continuous(trans='log10') + ggplot2::theme_bw() + ggplot2::ylab('number of cells')
    pl <- c(pl, list(szp))
  }

  pp <- cowplot::plot_grid(plotlist=pl ,ncol=1, rel_heights=c(1,rep(0.3,length(pl)-1)))
  pp2 <- cowplot::plot_grid(leg, pp, ncol=1, rel_heights=c(legend.height,1))

  return(pp2)
}

I would recommend playing with this code above. What you are doing is a standard ggplot2 operation.

Run the above code line by line and you will be able make the modifications you need to plot by condition as opposed by sample.

But we don't have the bandwidth to do this for you, and this isn't a new function for Conos----it's an analysis that you the user should do.

[kelvinyevng]

Thank you. Is it the sample.factor function?

[evanbiederstedt]

The sample.factor argument might be want you need. But you might have to manually change the ggplot2 commands as well for your use case.

This is a decision you'll need to make. You might need to write some custom ggplot2 commands here, given the above. Please feel free to share your solution with everyone in this forum.