We wish to replace microarrays with RNA-seq..we've seen papers published that specify that technical quality is quite repeatable, and thus technical replicates are unnecessary. We have also seen publications that have used RNA-seq with no biological replicates to generate results showing gene signature differences. If we are to replace microarrays with RNA-seq, is it bad practice (I know there's publications, but still...) to run the exoeriment without biological replicates? It becomes much more expensive if we had to run these in duplicate, so it's a huge difference, and we wish to know how we should proceed and what things we need to think about here. Any input would be greatly appreciated. Thank you very much.
You DO need biological replicates. Especially if the goal is similar to the one for microarray.
When you see that across two treatment, gene X has twice as much RNA molecules, how do you know if that is effect of random variability or an actual effect of the treatment? The only way is to measure variability within the same treatment/condition and compare it to variability across different treatment/condition. That means you need biological replicates and statistical tests.
Sequencing is somehow better than microarray because the actual number of sequences give also a confidence interval of the "real" expression level for that biological sample, but in no way it can tell you about the mean of a population of biological samples (that is what you are after)
Technical replicates are by far less relevant. If you think there might be some technical bias, randomize it with the biological replicates (if you think that the kind of "TAG" might have an effect, do not tag all normal with the same tag, but mix them)
I understand that cost is an issue, but with high throughput sequencing you should be able to tag RNA of different biological samples and run them together. You are always in time to "join" experiments if, for some reason, you don't care about variability among biological replicates. The opposite is impossible.
Also in reference to how the use of no replicates was seen in the literature, that doesn't make it okay. Just because someone may publish without replicates doesn't make it okay for everyone. It may have just been in that instance with that application it was okay. Or maybe it wasn't and the authors got away with something. I've seen reviews on SOLiD that stated only one replicates (two total) are necessary, but never no replicates.
Just to be clear, replacing microarrays with sequencing does not mean replacing the experimental design with another inferior experimental design. Any experiment about which you want to make an inference (think statistical test) needs replication. If cost will prohibit a good experiment, consider sticking with microarrays for the time being.
If you just want to "replace" microarrays with RNA-seq, and are happy with just reasonable gene-locus-level estimates of expression than you maybe be able to multiplex a significant number of samples together, bringing down your costs, and allowing sufficient replicates for well-powered studies. If however, you are hoping to get something more or different from your RNA-seq experiments than from microarrays (inferences about splicing, detection of mutations, allele-specific ratios, etc) then you will want more depth of sequencing. In that case, you will multiplex less (or not at all) and you may be able to live with less replicates in order to get at those other kinds of events. In an ideal world, we want both lots of depth and biological replicates. Fortunately, the plummeting cost of sequencing will soon make this feasible.
Hansen, Kasper D, Zhijin Wu, Rafael A Irizarry, and Jeffrey T Leek. “Sequencing Technology Does Not Eliminate Biological Variability.” Nature Biotechnology 29, no. 7 (2011): 572–573.
Hi All I have two biological replicates (that's what we can afford at this moment). how do I justify my statistical calculations with edgeR ( seems it does calculation with two replicates but reviewer is arguing with that). can somebody suggest something.
I'm guessing the reviewer feels like the experiment may not have been powerful enough. Assuming the two replicates within the control and experimental groups cluster together and there is a reasonable difference between the groups, what you could consider is making a list of interesting targets from among the results of your differential analysis and validating these using a cheaper method like PCR in a greater number of biological replicates.
Also in reference to how the use of no replicates was seen in the literature, that doesn't make it okay. Just because someone may publish without replicates doesn't make it okay for everyone. It may have just been in that instance with that application it was okay. Or maybe it wasn't and the authors got away with something. I've seen reviews on SOLiD that stated only one replicates (two total) are necessary, but never no replicates.