Question

Is there a way to get the confidence interval for the fold change generated on DESeq2

0

Entering edit mode

adeler001 • 0

@adeler001-21743

Last seen 2.8 years ago

Canada

The DESeq2 results table provides the fold change for each gene, but is there a way to get the confidence interval for each fold change the program generates ?

DESeq2 • 1.4k views

ADD COMMENT • link 3.0 years ago adeler001 • 0

score 0 · Answer 1 · 2021-11-09

0

Entering edit mode

James W. MacDonald 67k

@james-w-macdonald-5106

Last seen 2 days ago

United States

The output from results gives you the β and the standard error for the β, right? The confidence interval is just $\hat{\beta} \pm 1.96 \cdot \mbox{SE} \left( \hat{\beta} \right)$, so you already have everything you need.

ADD COMMENT • link 3.0 years ago James W. MacDonald 67k

0

Entering edit mode

Uh, that didn't work. I expected markdown symbols to work... Anyway, the CI is just β ± 1.96 x SE. And you have both the β and the SE

ADD REPLY • link 3.0 years ago James W. MacDonald 67k

0

Entering edit mode

what do you mean by β ?

where did you get the 1.96 from ?

ADD REPLY • link 3.0 years ago adeler001 • 0

0

Entering edit mode

In this context β is the log fold change (it's the estimate of interest from your linear model, labeled as log2FoldChange in the output of results. And the SE is labeled lfcSE. And the stat column is the ratio of the two, and is a Wald test). And since it's a Wald test, and assumes normality, 1.96 is the 97.5th percentile of a normal distribution. I am presuming by CI you mean a 95% CI.

Put a different way, I got 1.96 from the first intro to stats class I took. If you don't already know something like that, it should be a big red flag to you. Analyzing these data is not the trickiest thing in the world (I do it, so how hard can it be?), but it's not the easiest thing either. You do have to know a bit about what you are doing if you want some assurance you are doing it right. So if you are missing some of the fundamentals I would recommend either filling in those blanks, or finding a local statistician to help (or at least to consult with).