RSEM Downstream Analysis

Question

[Anushka]

This question is similar to this. I have RSEM output files namely "quatify.genes.result" and "quatify.isoform.result". My question is reagarding the downstream analysis, I want to get differential expressed genes. the quatify.genes.result file has two columns one expected counts and FPKM value. How to convert these expected counts to raw counts, so that I can use DESEQ or DEGseq. Could anyone suggest me other other softwares for the downstream analysis on RSEM output/ or what could be a better approach to this.

Thanks

[Anushka]

It was FPKM and not RPKM...I am going to edit this in my question

Answer 1

[Devon Ryan]

You can't convert expected counts to raw counts, that's impossible. You should not use DESeq(2) or DEGseq to analyze this dataset. Limma (via voom()) would work and edgeR is apparently also OK with expected counts.

[Michael Love]

my two cents: DESeq2 on rounded estimated counts will behave similarly to edgeR on estimated counts. If the ambiguity of reads across genes is minimal, then estimated counts will behave similar to raw counts. However, where this can potentially break down (the Var(count) >= E(count) assumption) is with many ambiguous reads across genes.

[Devon Ryan]

I've been wondering what your opinion was on this. Have you thought about relaxing/removing the check for integers when a DESeqDataSet is constructed? I suppose the biggest concern is people shooting themselves in the foot by then inputting FPKM/RPKM...

[Michael Love]

No plans to relax the check. It helps us a lot.

[Anushka]

Thank you so much.