EdgeR DE results of two treatments questions

[lran2008]

Hi all,

I am using edger to find DE genes among two time points (P7, P28) compared with control. In my experiment, 6 cows were subjected to treatA and 5 cows were subjected to treatB. I have the miRNA seq data for control, P7 and P28 of each cow.

Now My Question is that:
The first time I got the DE genes of treatA (P28 vs control, P7 vs control) by just importing the seq data of the 6 cows in treatA to EdgeR. (I perform the same work for treat B separately.)

The second time I imported all the seq data of the two treatments together into Edger and I still could get the DE genes for treat A and treat B. (In this way, I can compare the effects of the two treatments)

HOWEVER, the results from the two approaches for the same treatment are different. I understand that the normalization factor changed when the data of the two treatments were put together.

BUT which one is more reliable? Should I analyze the DE genes of two treatments separately or together?

[dpryan]

Well, the proper comparison would be to compare one of the treatments at the various timepoints to an appropriate sham treatment at the same time points, but that train has pretty much left the station (i.e., it's too late to complain about that since you already did the experiment).

In real biological situations, there are typically changes overtime both due to a treatment and in the absence of one (biology is messy). In the first case, you're picking up changes that simply happen randomly over time and those that happen due to the treatment. In the second comparison, you somewhat account for random changes over time, but not completely because treatment B can cause its own host of time-dependent changes. Lacking any information about the actual treatments, I would predict that the second comparison is more reliable. In the future, though, you should talk to someone who's done animal experiments like this prior to doing them so you can setup the proper controls...your life will then be much easier.

[mikep]

if you use the GLM functions in edgeR then you can set your experiment up as the 2-way repeated measures anova that it actually is.

Section 3.3 of the edgeR users guide covers this.

[lran2008]

if you use the GLM functions in edgeR then you can set your experiment up as the 2-way repeated measures anova that it actually is.

Section 3.3 of the edgeR users guide covers this.

Thanks. This is exactly what I need.