[Bioc-devel] Does it matter whether I pool technical replicates for edgeR

[Kasper Daniel Hansen]

On Thu, Dec 6, 2012 at 2:29 PM, Ryan C. Thompson <rct at thompsonclan.org> wrote:
> Hi all,
>
> I'm working with an RNA-seq dataset where every biological sample has two
> technical replicates. Is important for me to merge the technical replicates
> so that my count matrix has exactly one column per biological sample, or is
> it ok to leave them separate? My worry would be that leaving them separate
> might make edgeR think that the technical replicates are biological
> replicates, leading to an underestimate of dispersion and overestiamte of
> significance.
>
> So does it matter whether I pool technical replicates (by simply adding
> their counts together)? Is edgeR expected to give different results
> depending on whether or not I pool?

Yes and yes.

As you say yourself, some of your samples are technical replicates and
some are biological.  edgeR (and any other statistical packages) needs
to know which is which, and you should be using a method that can
differentiate between the two sources of variation.  If you're using a
package that cannot differentiate (and I am a bit out of touch, so I
don't know if the current codebase of edgeR allows you to do so, but I
am sure we will hear from people more familiar with the code), you are
better off by summing the technical replicates.  However, if properly
used, tech reps allows you to quantify technical variability and
biological variability which can be important (although underused in
may bio papers).

Remember that the software and statistical model does not magically
figure out how your samples are related.

Small pet peeve: there are many types of technical reps, from using
the same library in different lanes of a flowcell (always
uninteresting - or at least the interesting aspects of this level of
variation has been addressed well in the literature -  but very
common), to actually doing two different library preps for the same
rna sample (much more interesting, and people sometimes do this) to
actually obtaining two different samples from the same tissue and
person (almost never done, but by far the most interesting level of
technical variation in my opinion).

Best,
Kasper

> -Ryan Thompson
>
> _______________________________________________
> Bioc-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/bioc-devel