DESeq2: bad experimental design and "not full rank" problems

[Erinyes]

Hi all.

In my lab, we have some old RNA-seq data that were produced by other users in the past. The experimental design was completely wrong, but I would like to analyze them in order to decide whether I should repeat the experiments or drop the whole idea altogether. I have already tried a few things, but I am afraid that all I see is technical noise.

Sample	Condition	Prep	Treatment
1	KD	A	shRNA
2	KD	A	shRNA
3	KD	A	shRNA
4	ctrl	B	shRNA
5	ctrl	B	shRNA
6	ctrl	B	shRNA
7	ctrl	B	shRNA
8	wt	C	no
9	wt	C	no
10	wt	C	no
11	wt	C	no
12	wt	C	no
13	wt	C	no
14	wt	C	no

I have performed DE analysis of KD vs ctrl and I got 1200 DE genes. Using contrasts, I have check wt vs ctrl and I got 70 DE genes. However, the general profile of both comparisons seems very similar (through heatmaps). Is there a way to correct for the different preps? I have seen in my recent data that the batch effect can be huge, but is there a way to extract some meaningful information from this set of data? Or it’s not worth trying to make sense of them at all?

When I tried ~ Treatment*Condition or ~ Prep + Condition, I got the error that the model matrix is not full rank (obviously).

Any suggestions will be greatly appreciated, or any experiences with problematic experimental designs!

Cheers

[dpryan]

If you happen to have a set of genes that you know a prioi to be constantly expressed regardless of treatment then you could use those for normalization and correcting the batch effect correction. If not, there's nothing you can do.

[Erinyes]

Yes, I was afraid of that.

I have already considered such a solution, but I do not really *know* of such genes. I can make some educated guesses but that's about it.

Thanks anyway.