Question about biological vs technical replicates in RNA-seq analysis

[bjackson]

Generally when I receive RNA seq data, I get biological and technical replicates. IE if there are two conditions (eg treated vs untreated), there will be three samples Treated-A, Treated-B, Treated-C, each with two technical replicates, Treated-A-Lane1 and Treated-A-Lane2.

What most of my group has previously done is just run them all as individual replicates (i.e. n=6). The only thing I could see this benefiting is to average out technical variation between the samples.

The other way I could think to do it is just automatically combining the technical replicates to increase the sequencing depth.

What do you all do? Use the technical replicates separately to increase n / average more variation or combine them to increase the depth?

[dpryan]

You can not use technical replicates to increase your N. If you do that, your results are simply wrong and any paper you publish with the results should be rejected. Your N is the number of biological replicates only.

Technical replicate are typically automatically combined (i.e., summed), since technical noise for (bulk) RNAseq is known to be approximately Poisson (this was shown in some of the early RNAseq papers, you can search for the reference). The only time people don't sum technical replicates is when they're comparing new methods, since then it's important to show that the new methods behave in a similar manner to the old ones.

[Bukowski]

I think you might be confusing actual technical replicates with your sequencing provider simply splitting a single pool across two lanes, which is desirable to remove batch effects between lanes - a better course than running some samples down one lane and some down another.

Devon is correct, they should be combined.