One-tailed test edgeR possible?
2
0
Entering edit mode
Sarah • 0
@2357cabb
Last seen 29 days ago
Germany

Is it possible to conduct a one-tailed differential expression analysis test using edgeR? I might have missed something in the documentation, but I cannot find anything definite.

I mean by this conducting a test where I only look for over expression in one sample group, but not in the other.

one-tailed edgeR DifferentialExpression test • 292 views
2
Entering edit mode
@gordon-smyth
Last seen 1 hour ago
WEHI, Melbourne, Australia

No, we don't provide one-sided tests in edgeR because we don't recommend them.

If you compare two groups which have roughly the same number of replicates and sequencing depth, then the one-sided p-value would just be half the two-sided p-value for genes where the logFC is in the same direction as the alternative hypothesis.

1
Entering edit mode
@james-w-macdonald-5106
Last seen 7 hours ago
United States

I don't believe it's possible to do so using edgeR, as neither a likelihood ratio test nor an F-test provides directional testing like that. You could use limma-voom or limma-trend and then simply re-calculate the p-values using pt with the correct argument for lower.tail, depending on the directionality of your comparison.

0
Entering edit mode

In limma, the one-sided p-value is just half the two-sided p-value for genes where the logFC is in the same direction as the alternative hypothesis.

0
Entering edit mode

Well yes, but that doesn't preclude one from using the Excel methodology of taking half the p-value and calling it a day.

0
Entering edit mode

Yes, that's what I'm saying. I was agreeing with your answer, just adding that it isn't necessary to rerun pt.