Question

DESeq2 lfcShrink and unshrunk fold changes

1

Entering edit mode

igor ▴ 50

@igor

Last seen 3 months ago

United States

In DESeq2, it's possible to keep the unshrunk fold changes in results() output using addMLE=TRUE, but that requires betaPrior=TRUE. This behavior seems to be based on the old workflow.

The more modern approach is to use lfcShrink(), but then the unshrunk fold changes are lost. Is there a way to run lfcShrink(), but somehow retain both types of fold changes? It's possible to take the fold change before and after lfcShrink(), but that seems not optimal.

deseq2 • 1.8k views

ADD COMMENT • link updated 6.1 years ago by Michael Love 43k • written 6.1 years ago by igor ▴ 50

score 3 · Accepted Answer · 2019-09-17

3

Entering edit mode

Michael Love 43k

@mikelove

Last seen 2 days ago

United States

You could do:

res <- results(dds, ...)

res2 <- lfcShrink(dds, ...)

Then combine res and res2 as you like.

ADD COMMENT • link 6.1 years ago Michael Love 43k

0

Entering edit mode

That's the option I use, but then you end up with two objects. I save the results object in case I need to access it in the future, so I was hoping to avoid having to save both of them. I could modify them, but then I am concerned there would be issues with having them in a custom format. It would great if lfcShrink kept the unshrunk fold change. Maybe that could be an option or would that potentially break something downstream?

ADD REPLY • link 6.1 years ago igor ▴ 50

1

Entering edit mode

How about:

res <- results(dds, ...)
res$lfcMLE <- res$log2FoldChange
res$log2FoldChange <- lfcShrink(dds, ...)$log2FoldChange

Seems easy enough for the user to do this manipulation, and then it's clear what is what in the table.

ADD REPLY • link 6.1 years ago Michael Love 43k

0

Entering edit mode

That is easy indeed. And that shouldn't break anything downstream?

ADD REPLY • link 6.1 years ago igor ▴ 50

1

Entering edit mode

I can't think of anything it would break.

ADD REPLY • link 6.1 years ago Michael Love 43k