How to get FPKM values for Ensembl

[rkawa]

Hello Everyone,

I am scratching my head because I just can't seem to find a way to calculate FPKM values for Ensembl genes. I aligned the reads by STAR, and got count data using HTseq.

For refSeq, FPKM calculation is relatively easy since there isn't much overlap in the genome. However, Ensembl genes contain so many isoforms and overlapping region is a problem in calculating FPKM.

I first thought getting unique exon regions in gtf files will do the work (this is in a different post). However, STAR aligner also aligns the reads with two (or more) spanning exons. Therefore, I have to take the unique exon junctions into consideration for FPKM calculation. So far I don't know how to do this effectively.

I wish HTseq had a tool to spit out all the regions that were used in read counting, then the problem can be solved easily.

If anyone has encountered and solved this problem, I will appreciate your thoughts and inputs.

Thank you!

RK

[dpryan]

It looks like you're trying to make things vastly more complicated than they are. If you just want FPKMs from the non-multimappers then just take the counts, divide by some gene length metric (there are many) and continue with FPKM calculation from there.

In general, though, I would just use Salmon or Kallisto, though they'll give you transcript-level metrics (just sum them to get gene-level values).

[blancha]

I wish HTseq had a tool to spit out all the regions that were used in read counting, then the problem can be solved easily.

featureCounts just does that, by default.
It computes the length of the genes from the GTF file used to count the reads.
As a bonus, it's much faster.

I take the gene lengths calculated by featureCounts from the GTF file, and then give the lengths in input to DESeq2's fpkm() function.

Or, you could use Cuffdiff, but that's a different pipeline altogether.