EdgeR to find DE and non-DE reads

[Jo Lee]

I’m using EdgeR to find differentially expressed reads within 100bp windows. To help my analysis later on I’d like EdgeR to give me an output that contains every 100bp window I have regardless of if it’s differentially expressed or not. I’ve attached log of what I normally do to get differentially expressed read output.

Any help on how to get the expression data back from all 2393345 windows would be appreciated.

Thanks

[dpryan]

Just give topTags a larger n= value.

[dariober]

Any help on how to get the expression data back from all 2393345 windows would be appreciated.

Hi- It seems you get only 77811 positions because you apply this filter:

keep<-rowSums(cpm(b)>1)>=2

(i.e. intervals where CPM > 1 in at least two libraries)

If you remove it (or put ">=0" instead of ">=2"), you should get back all the positions (not sure whether it would be a sensible and practical thing to do though).

Dario

[Jo Lee]

Yes I agree Dario that I may not want to remove the keep<-rowSums(cpm(b)>1)>=2 because it's still important for me to have intervals where CPM>1 in at least 2 libraries but everything else is zero. I'll try the way you suggest and compare my results to when I have 77811 positions.
Ryan is your suggestion to do
top<-topTags(et)
et<-exactTest(b)
topTags(et, n=2393345)

Or were you suggesting to change the n value for the topTags right at the end before I write to .csv? I've tried this before and it still returns just the filtered 77811 rows.

Hope we can keep nutting out this problem together.
Thanks.

[dpryan]

I guess I missed that the ~2.4 million sites were the prefiltering number (though, in hindsight, that makes complete sense), so my suggestion wouldn't be helpful. You'll need to do as Dario mentioned and just not filter if you want all of the sites. Usually there's a reason to filter things, but if you want to compare across experiments or something like that where the filtering will likely remove different sites in different experiments, then not filtering would make that process a bit easier