Problem with low count removal in edgeR ?

Question

[Björn]

I used following command to remove zero counts using edgeR package

y<-DGEList(data, genes = miRNAs_ID,group = diagnostics, remove.zeros = T)

which removed nearly 1000 miRNA (rows) from subsequent "y" that is used for analysis. Following command to exclude low counts was

keep<-rowSums(cpm>1)>=6
table(keep)

This left 600 miRNAs (rows). However, using commandline

cpm(y)[o[1:25],]

to list first 25 rows of miRNAs still shows those with zero counts. The above command was used after using threshold of CPM of 1. What is wrong with it ? I would suppose that the "y" should not contain any miRNAs with "0". Thanks

[Devon Ryan]

Show how you got cpm and an example of a row with all 0 counts.

[Björn]

I followed the paper literature

and after this step

o<-order(qlfGrAvsGrB$table$PValue)
cpm(y)[o[1:25],]

produces list of miRNAs "with zero" counts. See attached file. The row numbers are ID for miRNAs. while each column represents sample.

[Devon Ryan]

Those rows aren't all zeros.

[Björn]

My concern is why there is still zero counts (in some samples)?

[Devon Ryan]

Because they should be there. The remove.zeros option isn't doing what you think it is.

[Björn]

Is it really a problem after using

keep<-rowSums(cpm>1)>=6
table(keep)

[Devon Ryan]

Obviously it is.

[Björn]

I believe those with counts below 1 CPM (which includes 0) will be excluded in "keep" so why it is still a problem.

[Devon Ryan]

It's unclear if you're actually using keep or if >=6 is sufficient. As an aside, it's questionable whether you actually want to do this.