DESeq2: plotMA looks wrong
1
0
Entering edit mode
mdroog ▴ 10
@mdroog-13422
Last seen 6.0 years ago

I have single cell RNA seq data from two groups. When I use DESeq2 to explore differential regulation between the two groups, I visualize them in a plotMA graph. To my surprise this shows most genes are not around 0 for the y-axe (log2Fold) but -1. I added a link to the plot I made. What am I doing wrong?

https://imgur.com/a/C1ZRSAH

 

library(DESeq2)

# g1 = cell names group 1

# g2 = cell names group 2

# x = data frame with unique reads (raw) of all cells

x <- x[,c(g1,g2)]
x <- x[!grepl("ERCC",row.names(x))&!grepl("Rn45s", row.names(x)),]

des <- data.frame(row.names=colnames(x),
                  condition= c(rep("con1",length(g1)),
                               rep("con2",length(g2))))
dds <- DESeqDataSetFromMatrix(
  countData = round(x,0),
  colData = des,
  design = ~ condition)

dds <- DESeq(dds)
res <- results(dds)
plotMA(res, ylim = c(min(res$log2FoldChange, na.rm = TRUE), max(res$log2FoldChange, na.rm = TRUE)))

deseq2 • 1.1k views
ADD COMMENT
0
Entering edit mode
@wolfgang-huber-3550
Last seen 3 months ago
EMBL European Molecular Biology Laborat…

Why do you round the values (line countData = round(x,0))? DESeq2 works on the raw counts, which are integers. If you give it something else, then garbage-in / garbage-out.

Suppose you are feeding it actual read counts, then one possible explanation is that  the default DESeq "normalization" does not do the right thing for these data. You'll probably want to do more EDA to explore the data properties before fitting a complex statistical model. In the course of that EDA, you could try other normalization options, from within DESeq2 and beyond.

On a more general note, the raison d'être of DESeq2 is the shrinkage (of location and scale parameters), which is relevant for designed experiments with small sample sizes, such as your typical bulk RNA-Seq experiment. When comparing hundreds or thousands of cells, there is no need for shrinkage, and almost any old two-groups test should do.

Personally, I'd probably try (ZI)NB-WaVE.

Best wishes
      Wolfgang

 

ADD COMMENT

Login before adding your answer.

Traffic: 300 users visited in the last hour
Help About
FAQ
Access RSS
API
Stats

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.

Powered by the version 2.3.6