Question

DESeq2: How to correct for multiple comparisons

0

Entering edit mode

NG • 0

@ng-21603

Last seen 5.4 years ago

Hello,

I am relatively new to using DESeq2 so thank you in advanced for your help!

I have multiple samples (A, B, C, D, E) with three replicates of each sample. I would like to compare each sample to each other sample (e.g. A vs B, A vs C, A vs D, A vs E, B vs C, etc..). I am fairly comfortable with running each individual pairwise comparison and from my understanding, the adjusted P-values that I get as output are only adjusted for that individual comparison. I'm wondering how do I go about correcting for the 10 comparisons I want to make?

Here is the code I'm currently using.

ountData <- as.matrix((read.csv("counts.csv", row.names="gene_id")))
colData=data.frame(row.names = colnames(countData),sampleCondition=rep(c("E","E", "E", "F","F", "F", "G", "G", "G", "H", "H", "H", "I","I", "I" ,"J", "J", "J", "K", "K", "K")))
colData$sampleCondition <- relevel(colData$sampleCondition, ref = "E")
countData <- countData[, rownames(colData)]
all(rownames(colData) == colnames(countData))
dds <- DESeqDataSetFromMatrix(countData = countData,colData = colData,design = ~ sampleCondition)
dds <- DESeq(dds)
res <- results(dds, contrast = c("E","F","G" ,"H", "I", "K"))
res_e <- results(dds, pAdjustMethod = "BH", contrast = c("sampleCondition", "J","E"))
res_f <- results(dds, pAdjustMethod = "BH", contrast = c("sampleCondition", "J","F"))
res_g <- results(dds, pAdjustMethod = "BH", contrast = c("sampleCondition", "J","G"))
res_h <- results(dds, pAdjustMethod = "BH", contrast = c("sampleCondition", "J","H"))
res_i <- results(dds, pAdjustMethod = "BH", contrast = c("sampleCondition", "J","I"))
res_k <- results(dds, pAdjustMethod = "BH", contrast = c("sampleCondition", "J","K"))

Thank you! Noa

deseq2 • 3.6k views

ADD COMMENT • link updated 5.4 years ago by Michael Love 43k • written 5.4 years ago by NG • 0

0

Entering edit mode

Hi Noa,

I hope your problem was solved. Actually, I had the same question and just came across this post. I have 6 samples and I want to create a heatmap for a list of genes of interest for all these samples. To do that, I need to have normalized counts for all the samples but if I do pairwise comparison in DESeq2, I am not sure that would be OK to combine the values into one table to create a heatmap. I appreciate it if you share your experience on this. Thanks a lot, Hamid.

ADD REPLY • link 4.7 years ago Hamidreza Hashemi ▴ 20

score 0 · Answer 1 · 2019-08-13

(You can use 3x backticks to make code blocks on the support site)

I tend to list each contrast and describe that the FDR control is across genes. But that would imply that you should present the results from all contrasts. If you are going to test over many comparisons, and then selectively show only the significant comparisons, you may want to consider the stageR Bioconductor package for looking for any difference in any pair, followed by confirmation of which pairs per gene have a difference at the pair-wise level.