Meaningful DE using rpkm values in a RNA-Seq analysis

Question

[Seq225]

Is it conventional/required to use a cut-off value to make rpkm values meaningful? For example, in a transcriptome analysis if a contig/gene (mRNA) has rpkm values of 0.025 and 0.0025 in two different conditions, the DE is 10. The ratio is high but the actual values (0.025 vs 0.0025) are very low. Should I consider this DE as significant (with multiple reps and p-value <.05)?

Is there a guideline to follow/decide what is meaningful and what not?

Thanks!

Answer 1

[swbarnes2]

The ratio is high but the actual values (0.025 vs 0.0025) are very low.

No one can really understand what those values mean, because they are so corrected. If you had fanatically deep coverage, that difference might be perfectly real and valid.

Should I consider this DE as significant (with multiple reps and p-value <.05)?

Don't try to assign a p-value to differences in RPKM. Put the raw counts through DESeq or edgeR.

Answer 2

[c.chakraborty]

Use raw read counts for genes and put them in DESeq2 or EdgeR for analysis. With RPKM, your expression is normalized to the length of the transcript. Thus if you have a very long transcript, the normalized value will be lower. As it is a practice in RPKM based analysis, to use a cutoff for expression, for example 0.1 RPKM, hence there are chances that you will lose a lot many interesting transcripts. Therefore use the read count values generated by htseq counts or feature counts.