Slot to pull data from for DoHeatmap

Question

[bio_info]

I am carrying out some scRNA-seq analysis in Seurat and I have made a pseudobulk of my data, transformed the count data using round() (as the count data was not all integer) and run FindMarkers using DESeq2 and sorted by log2_FC. Now I want to visualise the output in the form of a hierarchial heatmap.

My question is, which slot do I pull the data from the Seurat object? If I use scale.data then I loose a lot of genes which are not scaled, if I use slots counts or data then the heatmap just looks weird and nothing looks differentially expressed. Does anyone have a solution to this problem?

Answer 1

[bk11]

I would use data from Seurat object and perform a Hierarchical Clustering using Complexheatmap.

library(Seurat)
library(SeuratData)
library(dplyr)
library(ComplexHeatmap)


data=UpdateSeuratObject(pbmc3k.final)
all.markers=FindAllMarkers(object=data)

all.markers %>%
  group_by(cluster) %>%
  dplyr::filter(avg_log2FC > 1 & p_val_adj < 0.05) %>%
  slice_head(n = 10) %>%
  ungroup() -> top10
head(top10)

mat<- data[["RNA"]]@data[c(top10$gene), ] %>% as.matrix()
mat[1:5, 1:5]

## scale the rows
mat<- t(scale(t(mat)))
mat[1:5, 1:5]

cluster_anno<- [e-mail]$seurat_annotations
quantile(mat, c(0.1, 0.95))

Seurat::PurpleAndYellow()
## make the black color map to 0. the yellow map to highest and the purle map to the lowest
col_fun = circlize::colorRamp2(c(-1, 0, 2), c("#FF00FF", "black", "#FFFF00"))

Heatmap(mat, name = "Expression",  
        column_split = factor(cluster_anno),
        cluster_columns = TRUE,
        show_column_dend = FALSE,
        cluster_column_slices = TRUE,
        column_title_gp = gpar(fontsize = 8),
        column_gap = unit(0.5, "mm"),
        cluster_rows = TRUE,
        show_row_dend = FALSE,
        col = col_fun,
        row_names_gp = gpar(fontsize = 4),
        column_title_rot = 90,
        top_annotation = HeatmapAnnotation(foo = anno_block(gp = gpar(fill = scales::hue_pal()(9)))),
        show_column_names = FALSE,
        use_raster = TRUE,
        raster_quality = 4)

[bio_info]

Thanks for the help, much appreciated!

[Ming Tommy Tang]

use the data slot, and as shown in the code, you then scale it by yourself.