I will be analyzing relative miRNA expression in cells treated in various ways using deep sequencing. I am not sure whether it is important to perform the deep-seq on 3 biological replicates for each condition (per one sequencing run), or two biological replicates would be sufficient to show relative changes in miRNA expression between samples. Please note that I will be doing the same thing with my negative control samples in the same sequencing run.
This is more of a technical question than purely bioinformatics or computational biology-related, but I'll give it a try.
We've had many such conversations in our group with (would be) collaborators - does s/he need more biological or more technical replicates? Which are more informative? Generally, our opinion is if you are very careful and rigorous in following the protocols, and if your sequencing machines are well tuned, technical replicates are not nearly as informative or necessary as are the biological replicates. There is likely to be much more variation in signal and signal:noise from the biological side than from the technical side.
My colleague has performed a similar experiment, going with 3 biological replicates. It was a bit small, he now thinks, because of the variation in miRNA signal. He'd go with more replicates next time. And this comes from a guy who helped design early Affymetrix platforms and is very, very good in the lab.
Go with more biological replicates and feel more confident in those miRNAs that are shown to be differentially expressed.
I would definitely say add biological replicates. I am doing an RNA-Seq study in cattle looking at mRNA expression. We have only one technical sample per condition but 6 biological samples (and the advantage of paired samples as well). The biological variability is very high; that's to be expected. Using edgeR for analysis I have been able to get very sensible regulated gene results.
I would agree completely with Larry_Parnell above that biological variability will drown out any technical variability. I've found exactly that - biological variability (e.g cow or timecourse effect) dominates when using MDS or other clustering techniques to examine the data.
If that was not the case then something is badly wrong somewhere and I'd have my suspicions about the data anyway.
Have a look at this:
Kasper D Hansen et al., “Sequencing technology does not eliminate biological variability,” Nature Biotechnology 29, no. 7 (2011): 572-573.
Hi everyone,
I am really interested in this topic. I am working as well on smallRNA sequencing data, and I have two biological replicates. I have also observed quite a large variation in miRNA expression data, and I was wondering if it's just my case or if it's an already observed phenomenon. There are not enough publication yet on deep sequencing data WITH replicates...
I was wondering which could be a good statistical tool (edgeR, Deseq? or what?) to analyze significant variation in smallRNA expression...
Do you have some experience?
Thanks,
erica
