Taqman and RNA-seq do not correlate

[DonDolowy]

Dear all,

some time ago we did Taqman for a couple of candidate genes on some human samples, that showed downregulation of these genes. We therefore decided to go for a low-depth RNA-seq on these 38 samples, but it turns out that the results are no near what we see with Taqman where we used verified probes from Life Technologies.

We are a bit lost now and dont know what to trust. I would appreciate any input! Thanks a lot.

Info about sequencing process:
19-plexing 2x50bp strand-specific on a HiSeq2500.

FastQC files showed great quality but a bit of contamination of index sequences, so I trimmed the reads with TrimGalore. Ran FastQC again and everything was great. I mapped to hg19 (UCSC from iGenome) using the Tophat 2.0.8. I get around 7-10 million mapped reads.
I then continued with Cufflinks, Cuffquant and Cuffnorm (without de novo discovery). I also tried using DESeq 2. Get about the same results (also independent on whether I use classic.fpkm, quartile og geometric normalization).

Thanks a lot - appreciate your input!

[mbblack]

When you did the qPCR, just how down-regulated were these genes? Where the differences in expression large or small?

RNA-seq is very problematic when looking at/for low level expression changes. The problem being that these will be low count genes in your mapped read set, and they will have very large variance (across your biological replicates). Those low count & high variance genes often will not pass typical thresholds (FDR or Fold Change cutoffs) for detection of differential expression.

In fact, in my own experience, conventional microarrays were much better at characterizing low level of expression change, while RNA-seq did perform better as expression levels increased (i.e. as feature count increased).