Is there a possibility of dividing a sample into two technical replicates if saturation is reached quickly in alpha-rarefaction? How would it be done?
I can’t imagine any reason why this would be a kosher thing to do, that would be like me doing a study with n=4, multiplying my results by 2 and reporting n=8.
And so you can imagine there is no plugin in QIIME 2 that readily does this. However, you can probably hack something by filtering to that one sample and rarefying to half of its total sequences, twice. Though, again, I would not recommend this at all.
Okey thank you. My biggest problem is that we only did one replicate per sample, because we make a pool with all the replicates trying to represent better all the experiment. And now I cannot do statitistic. I performed Alpha diversity with shannon, beta diversity with bray curtis PcOA and permanova with some groups of samples that have characteristics in common, a dbRDA and a UPGMA dendogram.
The problem is that I cannot performed a PERMANOVA between samples with only one replicate, and I cannot do a LefSE analysis to compare significatively abundance genera between samples.
Do you know other process to make it?
Can you clarify if you are referring to biological replicates, or technical replicates?
If technical, then it may not be a big deal, most microbiome studies I see do not use technical replicates, this is primarily a cost issue. But if you have pooled real biological replicates I’m afraid there is not much you can do now. Your signal has already been diluted and trying to somehow make technical replicates out of them is not statistically sensible, you are essentially removing biological variance from the equation which most of the statistical tests you would be working with are centered around that.
I am referring biological replicates
Yes I don´t know how can I continue really because the idea was write a paper, but without the possibility of made a good statistic…
Yes that is unfortunately a tough situation, I don’t have any good solutions for you here. Perhaps you can just show the data without statistics and use it as a ‘hypothesis forming’ pilot study? Always a good idea to consult a statistician before designing an experiment…