Question

Single sided P-value for DESeq2 Likelihood Ratio Test

0

Entering edit mode

glennli • 0

@636e91b3

Last seen 18 months ago

United States

Hi, Is it possible to obtain single-sided P-values when using DESeq2 to test ratios of ratios using the likelihood ratio test (ie for Ribosome profiling)? Is it valid to perform a Wald test using the log2FoldChange of the ratio of ratios (output from the LRT mode) from as the contrast, or is this flawed/statistically invalid?

dds <- DESeqDataSetFromMatrix(countData = as.matrix(count), colData = as.matrix(coldata),
                              design= ~ assay + condition + assay:condition)
dds <- DESeq(dds, test="LRT", reduced= ~ assay + condition)

wald <- DESeq(dds)
wald_results <- results(wald, contrast = list("assayrpf.conditionplus"), lfcThreshold = 0, altHypothesis = "greater")

Thanks for your help.

DESeq2 DESe • 1.2k views

ADD COMMENT • link updated 3.7 years ago by Michael Love 43k • written 3.7 years ago by glennli • 0

score 1 · Answer 1 · 2021-03-01

1

Entering edit mode

Michael Love 43k

@mikelove

Last seen 7 hours ago

United States

We only have single sided Wald tests. So you can drop the test="LRT" and reduced if you just need the one results() call above.

ADD COMMENT • link 3.7 years ago Michael Love 43k

0

Entering edit mode

Thanks for your reply Mike. Is the Wald test appropriate when testing ratios of ratios? What do I lose from applying this test as compared to the LRT?

ADD REPLY • link 3.7 years ago glennli • 0

1

Entering edit mode

It's appropriate yes. By the way you can just do results(dds, name="...") if you are pulling out one coefficient.

I think the LRT is sometimes more robust when there are scattering of 0's and some very large values, so you could always filter such that at least X number of samples have a moderate count.

ADD REPLY • link 3.7 years ago Michael Love 43k