correlation among biological replicates

[buthercup_ch]

Hello,

I am trying to analyze some data that I just got from RNA-Seq from gram-negative bacteria.

The bioinformatician just gave me 2 sets of tables 1) reads counts, 2) FPKM.
Sequences has been mapped using Tophat as paired-end (PE), and FPKM calculated using Cufflinks.

The question is that if I analyze the Pearson's correlation, the result I got by using either the reads counts or the FPKM is different. I calculated Pearson's correlations either manually or using R package (corrplot package):
> Mcor <- cor(M, method = c("pearson"))
> corrplot(Mcor, type = "upper", order = "original", tl.col = "black", tl.srt = 45)

The point is that I do not understand why the good correlation I got when using reads counts (0.93-1) is not kept when using FPKM (descrease up to 0.19) (see figure attached).

I have to say that I did the correlation analyses using the whole data, including CDSs, rRNAs, tRNAs and ncRNAs (a small contamination of these RNAs still present after RNA depletion, about 25%).

Can anyone help me to understand this?

Thanks in advance.

Best regards