Question

Compare effect size of gene (lfcShrink or not)

1

Entering edit mode

Yuya Liang ▴ 20

@yuya-liang-14202

Last seen 5.7 years ago

United States

Hello everyone,
I'm using DESeq2 to analyze gene count data. I'm wondering should I compare effect size based on log2foldchange with or without
lfcShrink.

This is my MA plot for without shrinking and with `lfcShrink`.

I know it's better to compare gene effect based on shrunken log2foldchange. But in my case, will you recommend me to shrink data (log2foldchange became almost -2 to 2)? My another concern is that I also compare interaction but this version of

lfcShrink(, type = "normal")

doesn't shrink interaction term. would it be better just use non-shrunken log2foldchange for all? (I'm looking forward to apeglm and ashr)

Thank you!

deseq2 rnaseq • 1.5k views

ADD COMMENT • link updated 5.9 years ago by Michael Love 41k • written 5.9 years ago by Yuya Liang ▴ 20

score 1 · Answer 1 · 2018-06-20

1

Entering edit mode

Michael Love 41k

@mikelove

Last seen 6 minutes ago

United States

Why not upgrade to the latest version of Bioconductor (e.g. DESeq2 version 1.20)? Then you can use type="apeglm". It works with interaction terms, has very little bias, and still gives low error effect sizes (the top ranked estimated LFC are likely to be the true top ranked LFCs). It's typically just as fast as type="normal".

ADD COMMENT • link 5.9 years ago Michael Love 41k

0

Entering edit mode

Hi Micheal,
Thank you for the reply! I didn't notice there's the updated version of DESeq2 package. I tried to update to ver 1.2 but somehow it didn't work.

> packageVersion("DESeq2")
[1] ‘1.16.1’

I tried both update.packages() and remove.packages()

source("https://bioconductor.org/biocLite.R")
biocLite("DESeq2")

ADD REPLY • link 5.9 years ago Yuya Liang ▴ 20

1

Entering edit mode

See the instructions here for updating Bioconductor (current version is 3.7, you have 3.5):

https://bioconductor.org/install

ADD REPLY • link 5.9 years ago Michael Love 41k

0

Entering edit mode

Thank you! It works.

ADD REPLY • link 5.9 years ago Yuya Liang ▴ 20

0

Entering edit mode

It seems not too much different. Is that normal?

This is the code I used:

DESeq.A <- DESeq(DESeq.ds, test = "Wald")
a <- results(DESeq.A)
resLFC <- lfcShrink(DESeq.A, coef = 2, res = a, type = "normal")
resLFC.ape <- lfcShrink(DESeq.A, coef = 2, res = a, type = "apeglm")
par(mfrow = c(1,3))
DESeq2::plotMA(a, ylim = c(-4, 4), main = "DESeq2")
DESeq2::plotMA(resLFC, ylim = c(-4, 4), main = "lfcShrink_normal")
DESeq2::plotMA(resLFC, ylim = c(-4, 4), main = "lfcShrink_apeglm")

I read the manual, ashr is in DESeq2 v. 1.2 but I got an error when I was installing the related package (REBayes):

> install.packages("https://cran.r-project.org/src/contrib/REBayes_1.3.tar.gz")
Warning in install.packages :
  package ‘https://cran.r-project.org/src/contrib/REBayes_1.3.tar.gz’ is not available (for R version 3.5.0)

ADD REPLY • link 5.9 years ago Yuya Liang ▴ 20

0

Entering edit mode

Check your code again. You’re not plotting the “ape”’object.

ADD REPLY • link 5.9 years ago Michael Love 41k

0

Entering edit mode

I saw it... Thank you!!

ADD REPLY • link 5.8 years ago Yuya Liang ▴ 20

This is my MA plot for without shrinking and with lfcShrink.

This is my MA plot for without shrinking and with `lfcShrink`.