Spearman correlation test: FPKM or counts?

[travelk]

I'm not sure if this is very obvious but I'm not a statistician so I thought I would confirm with the more knowledgeable people on this site.


I would like to do a Spearman correlation between genes on my RNA-Seq data (which I've seen done in various papers). However, they only say "Spearman correlation on gene expression data" and I don't know if that means on counts or FPKM. I've looked around and some papers do FPKM/RPKM (but not necessarily between genes but rather between conditions/treatments) but in message boards, they normally recommend counts but without explaining why. Can anyone enlighten me? Is there any difference between the two?

I've normalized the counts using DESeq size factors.

Thanks for your insight!

[swbarnes2]

RPKM is counts, but corrected for differences in gene length between genes, and differences in overall read numbers between samples. You certainly do need to correct for the fact that not all samples get the same # of reads; you don't want one sample ranking as #1 for every gene because it got twice as many reads as everything else. Correcting for gene length is likely not going to matter since you are ranking.