gdagstn / hexdimred

Hexagonal binning for dimensionality reduction plots

Hexagonal binning with transparency for single cell data

hexDimRed

Plots a hexbin version of a dimensional reduction representing values from a series of columns of colData. This is exactly what schex(https://github.com/SaskiaFreytag/schex) does, but with added graphical features afforded by ggplot.multistats (https://github.com/flying-sheep/ggplot.multistats) and a slightly more flexible approach.

1. for numeric variables, the fill aesthetic is given by a function that calculates summary statistics (e.g. mean, median, var) per bin. The function is specified by stat_fun_numeric and will be applied to each bin separately.

2. for categorical variables, the fill aesthetic can be given by either calculating the most represented factor level in each bin (stat_fun_numeric set to "maj") or the proportion of a single reference level against all others ("prop“). This is achieved by setting all reference levels to 1 and all non-reference levels to 0, then calculating the mean. The reference level is chosen as the most abundant one in the column. In the future I may implement a choice for the reference level, although it could be tedious for the user who has to set each level independently for each column of colData.

3. the alpha aesthetic is set independently by alpha_by and can be the number (length) of observations, or any other operation performed on each bin.

4. gene expression values can be plotted specifying the assay

5. the user can supply custom color scales

The output is a list of ggplot objects.

Usage:

This is the traditional PCA plot as delivered by scater:

p1 <- plotReducedDim(sce, dimred = "PCA", 
               colour_by = "individual"
              ) + labs(title = "PCA - by individual")
p2 <- plotReducedDim(sce, dimred = "PCA", 
               colour_by = "n_UMI"
              ) + labs(title = "PCA - by total UMI")
p3 <- plotReducedDim(sce, dimred = "PCA", 
               colour_by = "subset_mito_pct"
              ) + labs(title = "PCA - by % mitochondrial expression")
p4 <- plotReducedDim(sce, dimred = "PCA", 
               colour_by = "scDblFinder.class"
              ) + labs(title = "PCA - by doublet class")

gridExtra::grid.arrange(grobs = list(p1, p2), ncol = 2)
gridExtra::grid.arrange(grobs = list(p3, p4), ncol = 2)

It is hard to make out the relative contribution of each level because of how dots are superimposed.

Using hexDimRed:

sce_hexdim <- hexDimRed(sce, 
                       stat_fun_categorical = "prop", 
                       other_fields = c("individual", 
                                        "subset_mito_pct", 
                                        "scDblFinder.class"
                                       )
                      )

sce_hexdim2 <- hexDimRed(sce, 
                        stat_fun_categorical = "maj", 
                        other_fields = c("individual", 
                                         "scDblFinder.class",
                                         "batch"
                                        )
                       )


gridExtra::grid.arrange(grobs = sce_hexdim[c(1,4)], ncol = 2)
gridExtra::grid.arrange(grobs = c(sce_hexdim[4], sce_hexdim2[2]), ncol = 2)
gridExtra::grid.arrange(grobs = sce_hexdim2[c(3,4)], ncol = 2)

As of June 17 2020, there are some additional functionalities:

• if there is no n_UMI column in the object, a simple density of cells in the bin will be plotted
• support for user supplied scales, both continuous (custom_cont_cols) and discrete/categorical (custom_cat_cols). Importantly, for the categorical variables, the color vector order will be matched with the factor order specified in other_fields using stat_fun_categorical = "maj".
• gene expression can be plotted via plot_genes, supplying the expression assay (default is lognormcounts)
• labels can be added with 2 styles: sticker (rounded box with solid background) or text (text only with no box)
• dark theme enabled

sce_hexdim3 <- hexDimRed(sce, 
          main_field = NULL,
          other_fields = "clustering_result", 
          stat_fun_categorical = "maj", 
          dimred = "UMAP", 
          nbins = 60,
          labels = "sticker",
          custom_cat_cols = custom.colors,
          plot_genes = "ENSG00000198963")
          
sce_hexdim3[[1]]
sce_hexdim3[[2]]
sce_hexdim3[[3]]

Dark mode:

sce_hexdim4 <- hexDimRed(sce, 
          main_field = NULL,
          other_fields = "clustering_result", 
          stat_fun_categorical = "maj", 
          dimred = "UMAP", 
          nbins = 60,
          labels = "sticker",
          custom_cat_cols = custom.colors
          dark = TRUE)
          

sce_hexdim4[[2]]