Deseq High Within Group Dispersion

Question

[learnerforever]

Hi guys,

I have posted a question and images on seqanswers

http://seqanswers.com/forums/showthread.php?t=21723

and was hoping someone here might have a few ideas.

Thanks!

Answer 1

[Jeremy Leipzig]

please post your the parameters you gave DESeq or, preferably, the whole R script

Did you try setting a lower count threshold, so you are not wasting multiple tests on contigs with one or two hits?

Answer 2

[learnerforever]

Thanks, Jeremy.

Size factors for the raw read counts were: 2.08 1.16 0.59 1.36 0.73 1.37 1.29 1.19 0.02 0.95 1.28 1.31 0.55 1.97 1.57 2.08 2.26 1.47 0.78 1.53 0.76 0.95

After size factor normalization, I got rid of low counts at 40% (use <- (rs > quantile(rs, 0.4)) as shown in the manual. Then I plotted the dispersion plots, that are showing high within group dispersion (~1). I am thinking that this high dispersion is generating very few diff genes (at p < 0.05, FDR 0.1) as between group variance must be very high to be called significant in the light of high within group dispersion. There is also a sample with very low counts (0.02 above) that I am thinking about getting rid of.