count based or cpm based filtering for miRNA normalization when library size is small.

Question

[Björn]

Hi, I used command

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

to filter zero counts and then used two filtering aproach

keep<-rowSums(y$counts) >500

gave ![enter image description here][1] [1]: https://ibb.co/kSk0nc and second approach

keep<-rowSums(cpm>1)>=3

gave figure ![enter image description here][1] [1]: https://ibb.co/e6ymSc

which one should I use before proceeding to downstream analysis Additional information : minimum lib.size is 1768648 while maximum is 5856985

Thanks

Answer 1

[ozarkp]

You should filter out low counts using the CPM. This is necessary because the raw counts for each gene should be normalized the library depth and the composition between samples. Otherwise, comparing gene expression between samples will be inaccurate and biased toward favoring longer genes or samples with more depth.