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

[Simon Anders]

Dear Lucia

On 18/05/13 14:01, Lucia wrote:
> To obtain my count matrix I use each Ucsc gene model as one "
> transcript", I limit my comparisons with the microarray data to those
> Ucsc gene models that have a unique RefSeq match with an Affy probe
> set

You have to understand that with such piecewise information, this thread 
is not very helpful for anybody.

So, for the record and for others who might read this thread: Lucia's 
observation that RNA-Seq shows less sensitivity than microarrays is 
unusual and not what one usually sees. The reason is likely either 
extraordinarily low read counts in the RNA-Seq experiment or a mistake 
in the generation of the count table.

Lucia: If you want the help of the list to get to the bottom of it, you 
have to explain what you did. This means giving a precise and complete 
description of your workflow, including all commands, scripts, 
parameters, outputs etc.

I don't understand why you don't want to tell us _how_ you created your 
count table (which command-line tool, R package, or self-made script 
have you used and how?), but without that, and some useful statistics 
about read counts (number of total and mapped read counts, column sum of 
count matrix, size factors reported by DESeq), this discussion won't 
lead anywhere.

> Regarding multiple testing, I guess there is no reason why I can't
> use the uncorrected pvalues produced by DESeq and run locfdr right?

Sure, this is fine, even though the use of _local_ FDRs is a it unusual 
in transcriptomics, where people prefer to use tail-based FDRs such as 
those reported by the Benjamini-Hochberg method.

   Simon