Cell attribution to one or another lineage about slingshot HOT 3 CLOSED

Hm, that is little strange.
One thing I can say is that Slingshot treats lineage assignment as a continuous quantity, so each cell has a weight associated with each lineage and the NA values in pseudotime just reflect weights of zero. For the cells assigned to both lineages that look like they should only be on one, what are their weights? If they are relatively small for the more distant lineage, this could potentially be solved by setting a stricter cutoff for inclusion on a lineage (such as a weight >= 0.5), making a few more NAs in the final pseudotime variables.

from slingshot.

Hello!

I have updated Slingshot as I saw you made modifications on the code.

I created my own function to retrieve the pseudotime for each lineage from the lambda, as it is not yet implemented in the new version (or I missed something...):

get_pseudotime <- function(pseudotime, wthres=wthres){
	pseudoT <- list()
	for(lineage in 1:length(pseudotime@curves))local({
		curve <- pseudotime@curves[[lineage]]
		lambda <- curve$lambda
		weight <- curve$w
		ps <- curve$lambda
		ps[weight < wthres] <- NA
		pseudoT[[lineage]] <<- ps
	})
	df <- t(do.call("rbind",pseudoT))
	colnames(df) <- names(pseudotime@curves)
	return(df)
}

When I set the weight threshold to weight >= 0.9, I have a much neater lineage that fit perfectly the biology (and my previous results with other methods).

And here is my code to plot single gene expression per lineage if you are interested. It is not very optimised but at least it works for me:

plot_gene_per_lineage <- function(rpkm_matrix=rpkm_matrix, pseudotime=pseudotime, gene=gene, condition=condition){
	myplots <- list()
	total_pseudotime <- vector()
	for (lineages in 1:ncol(pseudotime)){
		lineage <- as.vector(pseudotime[,lineages])
		total_pseudotime <- c(total_pseudotime, lineage)
		total_pseudotime <- na.omit(total_pseudotime)
	}
	max_exp <- max(log(rpkm_matrix[rownames(rpkm_matrix) %in% gene,]+1))
	max_pseudotime <- max(total_pseudotime)
	for (lineages in 1:ncol(pseudotime))local({
		i <- lineages
		lineage <- pseudotime[,lineages]
		sub_data <- log(rpkm_matrix[rownames(rpkm_matrix) %in% gene,]+1)
		data <- data.frame(
			pseudotime=lineage,
			gene=t(sub_data),
			condition=condition
			)
		colnames(data) <- c("pseudotime", "gene", "condition")
		data <- na.omit(data)
		p <- ggplot(data, aes(x=pseudotime, y=gene))+
		geom_point(shape = 21, size = 2.5, stroke=0.5, aes(fill=condition))+
		geom_smooth(color="black")+
		# xlim(0, max_pseudotime)+
		ylim(0, max_exp)+
		theme_bw()
		myplots[[i]] <<- p
	})
	plot_grid(plotlist =myplots, nrow=ncol(pseudotime))
}

from slingshot.

Hi!

Sorry for changing so many things the other day (I'm making a big push to get the package ready for submission to Bioconductor), but I really appreciate all the feedback!

There is an accessor function for pulling out the matrix of pseudotimes (called pseudotime) as well as the matrix of weights (curveWeights). So for setting a weight threshold, I think this might be a little easier:

pst <- pseudotime(slingshot_output)
pst[curveWeights(slingshot_output) <= wthresh] <- NA

And thanks for individual gene plotting suggestion! Now that the big overhaul is done, that's the next thing I hope to add.

from slingshot.