[BioC] [Bioc] RNAseq less sensitive than microarrays? Is it a statistical issue?

[Wolfgang Huber]

Dear Lucia

I just googled one of the words I had used and realised it has a much more sinister meaning than I ever intended. My apologies for that very unfortunate choice. A spectacularly incompetent attempt at using informal/slang language in a place where it does not belong.

You reported a large difference between the results of the 'locfdr' and 'BH' methods for multiple testing correction on the microarray data. This could happen (among a few other reasons) if there are problems with data quality, such as batch effects, that affect the distribution of the test statistic for non differentially expressed genes. How does the histogram of unadjusted p-values look like? For the BH-method, it should look like a mixture of a uniform between 0 and 1 (for the true nulls) and a peak at the left. See also, e.g., Fig 3c in http://openi.nlm.nih.gov/detailedresult.php?img=2865860_btq118f3&req=4

The bias towards up-regulation in the RNA-Seq data is curious. It does not seem to be caused by unequal library sizes. One way to get to the bottom of this could be the MA-plots - how do they look like?

Best wishes
	Wolfgang

On 21 May 2013, at 23:19, Lucia Peixoto <luciap at iscb.org> wrote:

> Dear Simon,
> 
> My apologies for not being concise in explaining the problem. I am pretty
> new to the list and I can see that it can be frustrating to try to help
> someone that is not giving all the details you need to do so. I appreciate
> your apology, since some of the initial responses to my question had a very
> unfortunate choice of words. Going through the suggestions of the thread
> has been very helpful. So here is a more thorough description of what I
> did, as well as some observations made after taking into account
> suggestions: