Low Fpkm Values From Cufflinks ?

Question

[GouthamAtla]

I have aligned my data ( paired end RNA-SEQ ) to genome ( Hg 19 - ensemble ) using tophat2 and with a GTF file and default options. When I give the same referance GTF file and accepted_hits.bam to cufflinks ( v2.1.1 ), the isoform.fpkm_tracking file has FPKM values of range 1.66072e-316 for some of the transcripts. Is it normal ? Can we consider them as less abundant transcripts and move on with downstream analysis ?

cufflinks command is

cufflinks -o outdir -p 5 -G ref.gtf sample.bam

Answer 1

[Istvan Albert]

A number that small is bound to be an artifact of computation - for most intents and purposes the abundance of that is zero.

[Charles Warden]

I agree - in fact, low coverage genes tend to yield unreasonably high fold-change values, so I use log2(RPKM + 0.1) for analysis (although that is not to say values less than 0.1 are technically problematic).

Here is a paper with a more detailed explanation:

http://bioinfo.aizeonpublishers.net/content/2013/6/285-292.html

The paper also includes some benchmarks with other algorithms, which is emphasized more here:

http://cdwscience.blogspot.com/2013/11/rna-seq-differential-expression.html