differential expression analysis for low coverage datasets

[chouech]

Hi,

I have RNAseq data with 1.5 M reads. But the general transcriptome coverge is very low. The percentage of gene covered over 70% at a given expression level is 20% only!

So I was thinking that for differential expression analysis I'll get low power because I've got a lot of gene with low coverage that may be false positive...

Do you think it could be a good idea to keep for differential analysis only gene with a moderate coverage like at least 50%??

When I do this I can found more differentially expressed gene! (7 instead of 2)

If someone has an opinion on this...

[Brian Bushnell]

1) It would be helpful to know what organism you're using.
2) Regardless, 1.5M reads very low for a good DE experiment in anything larger than a bacteria (and low even for a bacteria). It would only be informative about a handful of genes. And your 'coverage' metric is less useful than knowing the total number of reads mapped to a gene.

It sounds to me like you should generate a lot more data if you want useful results.

[chouech]

It's a fungal genome: Magnaporthe (40 Mbp) but I can't sequence more data... so I was thinking to try to do with what I have analysing only genes that have enough mapped reads... the problem is to choose a threshold. That's why I was thinking to keep only gene that are 50% covered at least... but that's true that is some part of the gene that may represent only 1 read!