[BioC] Small RNA seq data analysis using DESeq

[Simon Anders]

Hi Vedran

 > I have a question regarding the analysis of small RNAseq data using 
DESeq.
 > While counting the reads per loci I have weighted the reads by the
 > reciprocal of the places to which the read maps.
 > I was wondering whether it is still proper to use the negative binomial
 > test implemented in DESeq (after rounding the expression estimates) to
 > determine which loci are differentially expressed?

No, for two reason:

1. DESeq expects raw counts. Your weighting violates the assumptions 
behind the nehative-binomial model.

2. Imagine two of your loci are quite similar, such that most reads that 
map to locus A also map to locus B. Further, imagine that you compare 
treated samples to control ones, and locus A gets upregulated in 
response to the treatment while locus B is unaffected. With your method 
of summerizing the data, all the additional reads that the upregulated 
locus A produces in the treatment samples will also be counted for locus 
B, and hence, you will wrongly conclude that both loci react to the 
treatment.

Note that the second issue is a problem not only to NB-based method, but 
rather shows that you approach is in general not suitable for 
differential expression analyses.

   Simon