Hello, I´m interested in finding reference genes for qPCR in my RNAseq data. I use therefore the following parameters: altHypothesis="lessAbs",lfcthreshold=0.5, alpha=0.05 to detect the most stable genes: I am not sure if I have understood this correctly: So the alternative and null hypothesis are swapped, i.e. if pval<0.05, the null hypothesis (=difference) is rejected and the alternative hypothesis (difference smaller than 0.5) is accepted.

1) Do I understand it correctly that the genes with the lowest p-value are the most stable genes when I extract my results? 2) When I ask for the summary(res), it says:

out of 67793 with nonzero total read count
adjusted p-value < 0.05
LFC > 0.50 (up)    : 0, 0%
LFC < -0.50 (down) : 621, 0.92%
outliers [1]       : 3331, 4.9%
low counts [2]     : 61080, 90%

I´m wondering, why the summary shows the number of genes greater than my threshold, when I´m more interested in less.

3) In the Beginners guide it says: lessAbs - |β|<x - p values are the maximum of the upper and lower tests<--- what does this mean?

I`m thankful for any suggestions!

hi,

Looking at the code, I think summary.DESeqResults doesn't take into account the direction of the test. So you can just get these tables by hand:

with(res, table(sign(log2FoldChange), padj < Y))

The p-values we construct are described in more detail in the 2014 paper if you want to look there. We construct a test that is conservative, but gives error control for the null hypothesis.

Finally, regarding whether these are biologically stable, unfortunately that depends on the normalization (scaling factors). It's possible that despite the way we set up the null hypothesis, if the scaling factors are wrong (due to global shifts in expression), then you wouldn't be capturing the stable features. With global shifts in expression, you need some additional information either from the experiment or from prior knowledge of the system.