how to deal with technical replicates

Hi all,

I am doing rumen microbiome analysis. And I would like to extract rumen fluid from cattle to do in vitro analysis. For rumen fluid collected from each of the cattle, I would like do compare the effect of treatment to the microbiome profiles and fermentation characteristics using in vitro culture. So I think rumen fluid from different cattle can be considered as biological replicates. For chemical data, I can simply average the technical replicates. My question is how do I deal with the technical replicates for microbiome data? Should I compare the similarity of the sample and choose only one of the replicates or should I merge them? or can I simply combine the DNA from each replicates as one sample for sequencing?

I have came across this post, however, they were talking about simply sequence the same sample for multiple times while in my case, the samples are actually from different replicates.


Hi, Ming!
Currently I am analyzing a rumen microbiome as well :wave:.
I am treating rumen fluid samples from different cows as biological replicates. I did not pool DNA in one sample, neither I merge them in my table. To perform any statistical analysis, or to get an overview of microbiome it is better to have replicates, rather than pool them together.

Hi @timanix, thank you for your reply.
I am not talking about merging biological replicates, I am talking about technical replicates. Say for rumen fluid collect from each cattle, I am doing in vitro culture and inhibition study. So rumen fluid from one cattle will be used in multiple in vitro culture as technical replicates.

Sorry, I misread your post.
Hope, this time I understood you correctly: You have biological replicates from different cows, plus technical replicates (rumen fluid from one cow divided for replicates in in vitro system). In such case, I would sequence each replicate (biological and technical) separately.
In our case, we sampled rumen fluids for sequencing before in vitro system, and mixed them (same breed). Mixed rumen fluids were replicated in in vitro system and sequenced separately.

Hi @timanix , that’s what I thought, but after gone through the post I mentioned in my post, I am not sure about what to do. They argued about the biases from technical replicates as microbiome analysis is sample size dependent. However, what they talked about is sequencing the same sample multiple times while mine is not exactly the “same” samples. They are from different in vitro cultures. For chemical data from technical replicates we usually average them for stats, but I am not sure how to deal with sequencing data.

In my opinion, all your replicates are in fact biological replicates, since they are sampled from different in vitro cultures, even if some of them are originated from one cow, and other from different cows. I would treat them accordingly, and you always will have an information about the origin of each sample (cow + in vitro culture) in your metadata file, so you will be able to merge this samples later if necessary.
It is different from technical replicates in the discussion, since in their case it were the same samples, sequenced multiple times.

Thanks @timanix , that makes sense. However, can you specify how would you like to deal with the variation among cows. Do you mean merged the biom table from in vitro replicates from same cows? For example, for constructing PCoA plot for beta-diversity, will you treat culture replicates as individual points (if you do so, how do you deal with the variation from cows), or merge them? For me, merge the table make more sense.


I would prefer not to merge at all. So, my answer on this question:

will be that I would keep them as separate points. You can differentiate cows by color, size or shape of points.
But, if it is necessary, or you want to clean mess in PCoA, you will be able to merge samples by any metadata column (for example, merge samples from one cow, as you described). Note that by merging you will decrease statistical power of the analysis, but it depends on your setup.

PS. In case (only) if there is a limit in the number of samples to sequence, I would also consider choosing representative samples from in-vitro replicates or pooling a DNA to one sample.

Thank you very much @timanix for your detail explanation.
Just to verify one last thing, I could treat all samples within same treatment from different cows as replicates when comparing control vs. treatment no matter microbiome data or chemical data as they are all the biological replicates. Is that correct?
Sorry, my stats knowledges is restricted to nutritional and chemical data. Normally, I would use mixed model and with RCBD to treat different cows as block. However, as microbiome data are usually non-normally distributed, my knowledge is limited regarding non normally distributed data.


I suppose so
For alpha and beta diversity metrics Kruskal Wallis and permanova tests are implemented in alpha and beta group significance plugins.