If scale=FALSE, DotPlot logs data such that we get negative values

[ktrns]

Hi there,

You use log1p to LogNorm counts. In the Dotplot function, you use the normalised data, and unlog it to compute an average:

    data.plot <- lapply(X = unique(x = data.features$id), FUN = function(ident) {
        data.use <- data.features[data.features$id == ident, 
            1:(ncol(x = data.features) - 1), drop = FALSE]
        avg.exp <- apply(X = data.use, MARGIN = 2, FUN = function(x) {
            return(mean(x = expm1(x = x)))
        })
        pct.exp <- apply(X = data.use, MARGIN = 2, FUN = PercentAbove, 
            threshold = 0)
        return(list(avg.exp = avg.exp, pct.exp = pct.exp))
    })

And then, in the case of scale=FALSE, you log the averaged data again:

   avg.exp.scaled <- sapply(X = unique(x = data.plot$features.plot), 
        FUN = function(x) {
            data.use <- data.plot[data.plot$features.plot == 
                x, "avg.exp"]
            if (scale) {
                data.use <- scale(x = data.use)
                data.use <- MinMax(data = data.use, min = col.min, 
                  max = col.max)
            }
            else {
                data.use <- log(x = data.use)
            }
            return(data.use)
        })

However, we end up with lots of negative values and this is misleading. Can you switch to using log1p here as well?

Thank you,
Katrin

[liu-xingliang]

Log scale implies negative values (original value < 1). Personally, from a software developer view, here for scale = FALSE, using log or log1p both make sense (of course, 0s in data.use should be taken care of). From a user view, I think a more informative legend title helps.

Waiting for development team, before that, a more add-hoc way could be get data.frame behind Seurat::DotPlot and re-generate using ggplot2::geom_point():

library(Seurat) # Seurat_3.1.5

dotplot.gg_data <- DotPlot(
    object,
    features = features
)$data

# Using scaled average expression (Z-score)
g.scaled <- ggplot(data = dotplot.gg_data, mapping = aes(x = id, y = features.plot)) +
    geom_point(mapping=aes(size = pct.exp, color = avg.exp.scaled)) +
    scale_radius(range = c(0, 6)) +
    labs(
         title = "Title",
         subtitle = "Sub-title",
         x = "Identity",
         y = "Feature"
     ) +
     guides(
         size = guide_legend(title = "Expression Percentage"),
         color = guide_colourbar(title = "Scaled Average Expression")
     ) +
     cowplot::theme_cowplot()

# log1p
g.unscaled_log1p <- ggplot(data = dotplot.gg_data, mapping = aes(x = id, y = features.plot)) +
    geom_point(mapping=aes(size = pct.exp, color = log1p(avg.exp))) +
    scale_radius(range = c(0, 6)) +
    labs(
         title = "Title",
         subtitle = "Sub-title",
         x = "Identity",
         y = "Feature"
     ) +
     guides(
         size = guide_legend(title = "Expression Percentage"),
         color = guide_colourbar(title = "Log(Average Expression + 1)")
     ) +
     cowplot::theme_cowplot()

# log
g.unscaled_log <- ggplot(data = dotplot.gg_data, mapping = aes(x = id, y = features.plot)) +
    geom_point(mapping=aes(size = pct.exp, color = log(avg.exp))) +
    scale_radius(range = c(0, 6)) +
    labs(
         title = "Title",
         subtitle = "Sub-title",
         x = "Identity",
         y = "Feature"
     ) +
     guides(
         size = guide_legend(title = "Expression Percentage"),
         color = guide_colourbar(title = "Log(Average Expression)")
     ) +
     cowplot::theme_cowplot()

[timoast]

This has now been fixed by @saketkc on the develop branch