Differentially Expressed Genes In Highly Correlated Samples?

Question

[Pfs]

I have an RNASEQ experiment with 6 samples (3 treatments and 3 corresponding controls, i.e. A,B,C and ctrlA, ctrlB, ctrlC).

I used cuffdiff to determine the differentially expressed genes/isoforms for the 3 comparison: A vs ctrlA, B vs ctrlB, C vs ctrlC and got some sets of interesting genes/isoforms.

When I compute the correlation between the samples and then clustering, I was expecting to see either the treatments clustering with their corresponding control or clustering together but separately from the control. Instead they all have high correlation > 0.98 with each other (treatment and control).

Is it possible to have DE genes among samples that are highly correlated? My first interpretation would be that the genes have a different expression magnitude across samples, but their trend is conserved across sample. Is this correct? What can I do to further test the data?

Thanks in advance

Answer 1

[Damian Kao]

It is possible to have DE genes among samples that are highly correlated. Some treatments just don't cause a wide-spread change in gene expression.

A short answer would be to do replicates to give you more statistical power on the DE genes.

A few things you can ask yourself:

• Was there a positive control in your treatments? I am assuming there is something interesting about this treatment that made you want to RNA-seq it in the first place. When you made your libraries, did you have a positive control to make sure your treatment worked?

• Do you have any genes or qPCR-ed genes that are known to be down/up regulated after the treatment? You can compare your results and see if they match.

• How did you map and calculate the expression levels? Did you use RPKM or tags for the DE analysis?

• Did you ribo-deplete the libraries? What is your estimated coverage of the transcriptome?

[Khader Shameer]

IMHO, a bit of formatting will further enhance the readability of this excellent answer.