Differential analysis heatmap without hclust

Question

[newbie]

Hello,

I did a differential analysis between 160 tumor and 90 normal samples. I'm using edgeR package. I used the following code to make a differential analysis heatmap. I selected differential expressed genes based on FC > 2 and FDR < 0.05. This gave me 614 DEGs.

logCPM <- cpm(y, prior.count=2, log=TRUE)
o <- order(tr$table$PValue)
logCPM <- logCPM[o[1:614],] 
logCPM <- t(scale(t(logCPM)))
dim(logCPM)

library(matrixStats)
library(gplots)
library(ComplexHeatmap)
library(circlize)
library(RColorBrewer)

#Set annotation
ann <- data.frame(TvsN$Type)
colnames(ann) <- c("Type")
colours <- list("Type"=c("Tumor"="black","Normal"="brown"))
colAnn <- HeatmapAnnotation(df=ann, which="col", col=colours, annotation_width=unit(c(1, 4), "cm"), gap=unit(1, "mm"))

myCol <- colorRampPalette(c("navyblue", "white", "red"))(100)
myBreaks <- seq(-2,2, length.out=100)
hmap <- Heatmap(logCPM, name = "Z-Score",  col = colorRamp2(myBreaks, myCol), 
                show_row_names = FALSE, show_column_names = FALSE, cluster_rows = TRUE,
                cluster_columns = TRUE, show_column_dend = FALSE, show_row_dend = TRUE,
                row_dend_reorder = TRUE, column_dend_reorder = TRUE, clustering_method_rows = "ward.D2",
                clustering_method_columns = "ward.D2", width = unit(100, "mm"),
                top_annotation=colAnn)
draw(hmap, heatmap_legend_side="left", annotation_legend_side="right")

The heatmap was made using complex heatmap like above. I see that DEGs showing samples are clustered.

![heatmap][1]

I wanted to select 614 DEGs and sort the samples acc to their Type without using hclust like above. May I know how to do this?

thanq

Answer 1

[Jean-Karim Heriche]

Read the docs about clustering in ComplexHeatmap. You can suppress clustering (and provide the data in the order you want) or give a custom clustering function or precomputed dendrogram.

[ATpoint]

In your case it would be cluster_rows which needs to be set to FALSE as by default it is TRUE and then uses Complete linkeage to cluster the rows. As Jean-Karim Heriche says, please read the manual, it is outstandingly comprehensive with an example plot for like every possible option.

Answer 2

[dtm2451]

You can do this in dittoSeq. dittoHeatmap() will even automatically grab the proper genes data subset and automaatically generate metadata annotations.

# Import full dataset from edgeR DGEList to the format dittoSeq understands
RNAdata <- importDittoBulk(your_DGEList_object)

# If your `type` data was stored within the DGEList, it might be imported automatically
# but to give example code if not...
# Adding the 'Type' metadata
RNAdata$Type <- types_data

# Make the heatmap
dittoHeatmap(
    object = RNAdata,
    genes = DEG.genes,
    annot.by = "Type",
    order.by = "Type",
    cluster_cols = FALSE)

Note on dittoSeq installation: If you are in R 4.0, you can install through Bioconductor with BiocManager::install("dittoSeq"). If not, you can still install it via the github (I have tested myself for R≥3.6.2, but this method should let you install in any R version.) BiocManager::install("dtm2451/dittoSeq")

Additional tweaks to the plot can be made by providing additional ?dittoHeatmap or ?pheatmap inputs, but the above code should accomplish the goals that you mention here.