Question

log fold change threshold in Deseq2 - using lfcthreshold

2

Entering edit mode

lirongrossmann ▴ 80

@lirongrossmann-13938

Last seen 4.2 years ago

Hi Everyone,

I am running a rna seq differential gene expression experiment with the following code:

dds <-DESeqDataSetFromMatrix(countData = ep,colData = cp,design =~Age+Gender+Response)
dds <- DESeq(dds)
res <- results(dds,contrast=c("Response","Yes","No"))

Looking at the results I get about thousands genes with absolute log fold change > 0.5.

However, when I run the following command:

resGA <- results(dds, lfcThreshold=0.5, altHypothesis="greaterAbs")

I only get 256 with absolute log fold change > 0.5.

Anyone can help me figure out what is the source of this discrepancy?

Thanks

lfcthreshold deseq2 • 6.0k views

ADD COMMENT • link updated 7.0 years ago by Michael Love 43k • written 7.0 years ago by lirongrossmann ▴ 80

2

Entering edit mode

Your first test is the following: pvalue is describing the probability that log fold change != 0, and the log fold change value is the most likely value given the data.

Your second test is: pvalue is describing the probability that the abs(log fold change value) < 0.5

You can therefore see how a gene with lfc of 0.51 may be significantly more than zero, but not significantly greater than 0.5

ADD REPLY • link 7.0 years ago kieran.mace ▴ 40

score 5 · Answer 1 · 2018-01-02

Using lfcThreshold is testing against an LFC threshold of 0.5, which is not the same as asking for which genes the estimated LFC is larger than 0.5. See the DESeq2 paper for discussion of this topic, under the heading, "Specifying minimum effect size":

https://genomebiology.biomedcentral.com/articles/10.1186/s13059-014-0550-8