Spike-ins in RNAseq...

[SylvainL]

Hello all,

I was wondering: how many of you (or researchers from your institutions) do add spike-ins in their RNAseq experiments? And in which context?

s.

[Brian Bushnell]

Perhaps you could clarify your objective? There are a lot of Illumina protocols that ask for a PhiX spike-in for any experiment; there are a lot of ribosomal RNA studies with low diversity that benefit from PhiX; and there are various molecular-weight spike-ins, cross-contamination-detection spike-ins, etc that can be useful in many scenarios. Given that - most, if not all, of JGI's RNA-seq experiments have some kind of spike-in at some point. But I don't think that's what you want to know.

[SylvainL]

Hi, no, I am not speaking about the spike ins during the sequencing step. I was wondering if any of you was asking to the researcher to add external RNA control (like the set of spike ins of the ERCC) to their experiment...
Until now, none of the researchers of my institution used them (so I do not have experience with these external spike ins) and I read different papers which conclude either, these spike ins are not good controls, either they can be useful... So I wanted to have the opinion of the SEQanswers community...

[blancha]

We never use the ERCC spike-ins.
They are an added cost and hassle of dubious benefit.

The main issue, confirmed independently in several studies, is that the reliability of the spike-ins themselves is questionable.
What is the purpose of a control that can be itself biased?

There are specific cases, such as when the total amount of RNA per cell varies between the conditions being compared, where the benefits of ERCC spike-ins may outweigh their disadvantages.

I would need a control on the control to start using ERCC spike-ins in run of-the-mill NGS experiments.. I just don't trust them, and believe that the data before correction is sometimes more accurate than the results after correcting with the ERCC spike-ins.

[jparsons]

I've spent a good deal of time working on the ERCCs and have a few points to make in their defense;

Yes, polyA selection biases affect some ERCC controls. There is no reason to believe this is different from endogenous RNA though.

Yes, RUV doesn't work well with ERCCs. They assume that a control spiked in at 1% of 2 different samples should always amount to 1% of the reads, when in fact spike-ins are added before rRNA depletion and rRNA depletion varies from experiment to experiment, and that 1% can easily become 2% due to the differential enrichment of spike-ins.

Now, to criticize the ERCCs a bit....There is very little public information of how to properly use them in sequencing, especially for normalization, and there are not a great many cases where the information they provide in sequencing is particularly actionable or useful.

However, if you look at cases such as the SEQC datasets where there are clearly-defined mixture proportions, and do classic deconvolution methods [ex: use CellMix] to attempt to solve them, those methods get the wrong answers (80/20 and 30/70 instead of 75/25 and 25/75). If you apply the ratio difference factor (Rm) reported by the ERCCdashboard on those same samples, the deconvolution proportions return to the correct values. I believe that the calculation of the differential total mass of mRNA in spiked samples is the most (and perhaps only) valuable information that ERCC spike-ins can provide.