RNA-seq: Strategy for filtering low count genes?

[LeonDK]

Hi all,

In the 2013 nature RNA-seq protocol by Anders S. et al. (http://www.nature.com/nprot/journal/....2013.099.html), they state: "In edgeR, it is recommended to remove features without at least 1 read per million in n of the samples, where n is the size of the smallest group of replicates"

In my case, this would be filtering on n=15 of a total of 96 samples.

Would you always apply low count-filtering regardless of DEG analysis method? (DESeq2, edgeR, limmaVoom) or would you simply include all genes and then afterwards look at the count values for any identified DEG?

Cheers,
Leon

[dpryan]

The most sensible thing to do is perform the tests and then perform independent filtering (see the genefilter package and the corresponding paper). This is done automatically in DESeq2, for what it's worth.