Question

DESeq2: group-group distance analysis (not only sample-sample distance)?

0

Entering edit mode

KimS • 0

@kims-13267

Last seen 6.9 years ago

Dear all,

I am new to DESeq2 and R and am in need of a distance analysis between different cell types in an RNAseq analysis. We have 10 cell types and 2 biological replicates for each. The sample-to-sample distance was easy to obtain with the code provided in the manual (thank you! and see below), but is there an easy way to assess the group-to-group distance too? (i.e. cell type to cell type). E.g. is it fair to average the counts of biological replicates in the dds, and if so, how? Is there a way to choose JS-distance over Euclidean?

rld <- rlogTransformation(dds, blind=TRUE)

distsRL <- dist(t(assay(rld)))
mat <- as.matrix(distsRL)
rownames(mat) <- colnames(mat) <- with(colData(dds),paste(celltype))

I truly thank you for any assistance that you can offer,

KS

deseq2 • 1.4k views

ADD COMMENT • link 6.9 years ago KimS • 0

0

Entering edit mode

Hi Michael,

Yes, this would absolutely work.

That you kindly for your effort in advance!

KS

ADD REPLY • link 6.9 years ago KimS • 0

score 0 · Answer 1 · 2017-06-15

0

Entering edit mode

Michael Love 41k

@mikelove

Last seen 2 hours ago

United States

How about the sum of squared shrunken log2 fold changes as a distance? I can write a script to calculate this for you.

ADD COMMENT • link 6.9 years ago Michael Love 41k

0

Entering edit mode

So you can do:

dds <- DESeq(dds, betaPrior=TRUE)
coefs <- coef(dds)[,-1]
dist.mat <- matrix(nrow=ncol(coefs),ncol=ncol(coefs))
diag(dist.mat) <- 1
for (i in 1:(ncol(coefs)-1)) {
  for (j in (i+1):ncol(coefs)) {
    dist.mat[j,i] <- dist.mat[i,j] <- sqrt(sum((coefs[,i] - coefs[,j])^2))
  }
}

ADD REPLY • link 6.9 years ago Michael Love 41k

0

Entering edit mode

Thank you, Michael.

When run, the all dist.mat values are NA, except the diagonal. The dimensions look good (10 by 10). Can you have another quick look at this and/or let me know how to help?

ADD REPLY • link 6.9 years ago KimS • 0

0

Entering edit mode

Put an na.rm=TRUE in the sum. The NA are from genes with all 0 counts.

ADD REPLY • link 6.9 years ago Michael Love 41k

0

Entering edit mode

Perfect! And thank you for your time!

ADD REPLY • link 6.9 years ago KimS • 0