Additonal replicates for RNA-seq from separate HiSeq run

[DRAT]

Hello,

We are planning to generate additional replicates for RNA-seq experiment and I would like to ask for opinions on the experimental design and make sure that our approach is valid.

We have experiment composed of 4 treatments with 2 replicates for each condition and the variation between replicates turned out to be pretty high. Samples come from primary mouse tissue so some biological variation can be expected but the main problem is that the pattern of read counts between samples and replicates doesn't match qPCR (done using the same starting material). For example for some genes that came out as differentially expressed we see that replicate B has many more counts, while in qPCR there's no difference or it's even opposite (I am comparing gene level counts generated with htseq and normalized using DESeq package to normalized qPCR expression).

I wonder if anyone has ever encountered similar problems when comparing RNA-seq counts to qPCR?

Also our replicates don't cluster together so we would like to generate two additional ones for each sample to have some more confidence in the results and find more differentially expressed genes. The problem is that new replicates will have to be sequenced in separate run and undergo separate library preaparation step. According to our sequencing facility the technical variation between the runs is very low but I would like ask for some more opinions on that. We would like to analyze samples from the first run together with new samples (2+2 replicates) and I wonder if anyone has ever tried it before and if it's allowed to do so?

Thank you for your comments,

[kopi-o]

In my experience the kind of technical variability that occurs from sequencing the same sample twice is indeed usually very low, but the technical variability from making a new library preparation (which is what you are doing, or did I misinterpret?) can be rather high and often needs to be taken into consideration as e.g. a covariate in edgeR or DESeq.

It sounds to me like you have a potential mix-up between samples, but who knows ...