Hi Steffie,

Yes, in my opinion the results of any global expression analytical approach should be validated by qPCR. The technique provides better estimates of expression levels of a particulat miRNA.. There are several well established approaches...such as stem loop PCR (where on has to do separate RT reactions for each miRNA) or some other approaches based on linker ligation or poly A polymerases for analysing miRNA from a pool of cDNA....go through some published literature..

We performed a poly A linker approach using a miScript cDNA synthesis kit. Then, we run a QPCR using  SybrGreen. But my question is, if anyone had difficulties before validating microRNA expression from deep sequencing data. As we have problems what to believe (deep seq or QPCR)

Yes it's always a problem when you have contradictory data of which to believe! I think the answer is to look in detail at both sets of data for clues as to which is more reliable.

I've never used the polyA linker approach so don't know how effective it is. I've always used the stem-loop primer approach which is expensive if you use the pre-designed TaqMan probes but pretty cheap if you design your own to use SYBR green, and has always worked well in our hands. (see our paper below for details if you're interested)

It would also be worth looking in detail at the RNA-Seq data. What is the calculated fold-change? If it's a small change you're looking for then how successful your validation is will be highly dependent on the quality of your qPCR data. In my experience qPCR is generally more sensitive than RNA-Seq, but the quality of qPCR data vary a lot from user to user. What is the standard deviation for your technical replicates, and your biological replicates? Basically, do you think you could identify the level of differential expression predicted by RNA-Seq with the level of variation you have in your qPCR data?

Also look at the levels in the individual replicates for the RNA-Seq and see how much they vary. If the analysis was performed in a pairwise manner, there may be consistent differences between the two conditions in each pair of replicates, but this would not be apparent by looking at the average levels of the two conditions across all replicates.

Are you performing qPCR on the same samples you used for RNA-Seq or different ones? Either would be interesting but would tell you different things. qPCR on exactly the same samples would be good to show you can technically replicate the results of the RNA-Seq. Once you've done this, qPCR on other biological replicates of the experiment would provide independent confirmation of any results.

So a final message as to which to believe? Look critically at both data sets but I probably wouldn't believe either if they didn't both give the same message.

Article miR-29a and miR-29b Contribute to Pancreatic -Cell-Specific ...

Thank you so much Timothy J Pullen for you extensive answer. I'll try to look at both data again with a critical eye and these comments in mind. FYI, we used the same samples for QPCR as for RNA-seq

I think one of the toughest things about validation of RNA deep sequencing is to admit the possibility that even though deep sequencing is a high-throughput approach, and thus traditionally thought of as less precise, it might actually be the most quantitatively reliable and may be most sensitive to small expression differences.  The various juries are still out on this.  Considering that a 1 PCR cycle difference = 2-fold abundance difference, and 0.5 PCR cycle difference = 1.4-fold difference, I feel that trying to detect anything less than a ~1.4-fold difference with qPCR is hazardous, considering how exacting the cycle differences will have to be.  On the other hand, there is often more precision (less variance) when counting sequencing reads, which can be important for identifying small differences.

Remember too that qPCR can be quite sensitive to the choice of reference genes.  Never use just one!  (and read papers by Jo Vandesompele on this if you don't believe me.) Again, I think an advantage of RNAseq approaches here is that each data point (miRNA or mRNA or whatever) is considered against the background of all other RNAs of the same type, in most of the common analytical paradigms.  I certainly think that qPCR validation is important to attempt, especially for key genes that may be critical to downstream experiments.  But don't drive yourself crazy if you can't get as precise an answer with qPCR as you do with RNASeq - in my view, it's the nature of the different methods.

By the way, hi Timothy; met you and Guy in Chicago for the ADA meeting a couple of years back!

Just a little add: I suppose you have to verify your validation protocols in order to see if you have your system working as you expected, for qPCR you can check on MIQUE Guidelines.