How to deal with techinical replicates in 16S amplicon sequencing data/OTU table?

Question

[cuicui]

I have 16S amplicon sequencing data and have two groups to compare. I used qiime2-deblur to get the OTU table. Some of my samples have technical 2 or 3 replicates. For example, for sample1, it has 2 technical replicates. How do I deal with these technical replicates? Can I sum or average them or normalized + average for the same OTU? Can I just use the OTU with the most count?

I appreciate it if someone could help me and give me some literature to learn.

Answer 1

[antonioggsousa]

Hi,

Assuming by technical replicates are just the same sample/tube sequenced independently twice, to assess the reproducibility of the library/sequencing itself, I would average them. I think if you sum them, the samples with 2 technical replicates will have less number of reads than the samples with 3 replicates. Although average them can also create additional problems because you can end up with float numbers, that are not ideal for a absolute count matrix. In that case I would probably round the values.

If you don't want to be in doubt, you can just assess by plotting the technical replicates if there is any difference between technical replicates or not. For this a PCA analysis can be a good idea. If the technical replicates look fine, i.e., there isn't any variation (at least not much) within technical replicates, you can just drop them for the downstream steps. Pick just one of the technical replicates and drop the remaining ones.

I hope this helps,

António

[cuicui]

Hi António, Very thankful for your answers! It is the same object, but different samples. We sampled different parts of the same object and sequenced them separately. And is it appropriate to call them technical replicates? Now I am a little bit confusing here. Maybe I could treat them as independent samples? Thanks again!

[antonioggsousa]

So, if they are different samples from the same object I guess these are biological replicates and not technical replicates. Technical replicates are replicates of the procedure/method. In this case it would correspond to extract DNA from one sample and, then divide this DNA from this sample in 2 or 3 tubes and sequence them separately to assess the variability of PCR + library preparation + sequencing on this sample. If you sample the same system/habitat/site multiple times these samples represent biological replicates to assess the biological variability of the system/habitat/site.

If you want to perform statistical analyses you will need biological replicates, although you might want for instance plot the taxonomic profile of microbial communities for one sample only, in that case average the counts, but yes you can treat them as independent samples. I hope this answers your question.

[cuicui]

Hi António, thanks again for this explanation! Here is the detailed description of the samples, samples are extracted separately from one the surface of the same stone. And I also have multiple stones. So samples are from different stones, they are definitely from separate biological samples. And samples from the same shell, but different surface(we only know they are from different parts, no other record information), they are also independent samples? Or they can both be treated as independent samples and at the same time, it can also be treated as average?

P.S. After your first answer, I also did a PCoA of different samples, most of the samples from the same stone, tends to cluster together, but I am not sure, which statistical methods I should use to evaluate? Does that mean there are some effects of the same stone?

I am truly thanks for your kind help, cuicui

[antonioggsousa]

Hi @cuicui,

All depends on your experimental design. Samples from the same stone or from different stones can be or not biological replicates.

Take these two examples:

(A) I'll sample/collect DNA from different types of stone in a desert. Let's say that I have 3 types of stones there - granite, basalt and limestone. In this place, I want to assess the difference between microbial communities across stones. Therefore, I'll sample different types of stones - biological samples -, and the same type of stone - biological replicates. In both cases I'll sample different stones.

(B) I'll sample the same kind of stone across some gradient, let's say a salinity gradient in a beach. Different stones will have different salinities, therefore, I'll sample the same kind of stone but different stones across the salinity gradient - biological samples -, and I'll sample also the same stone (so, the same salinity gradient) multiple times - biological replicates.

When I said independent what I meant was that you can leave them as they are. I mean you can plot them as biological replicates in taxonomic profile plots, PCoAs or alpha-diversity plots.

Independent and dependent samples in statistics means a slightly different thing: independent samples are samples sampled randomly from the population and dependent samples are samples that the same sample was exposed to different treatments, but there is a dependence on the sample because the outcome depends on the same sample (for instance drug and treatments on same individuals - the outcome of a drug-treatment will depend on the individual, because each drug-treatment combination was applied over the same/specific individuals).

If the biological variability in each stone is low, i.e., if the biological replicates are similar, you'll expect them to cluster together in a PCoA. That is actually good, because imagine that you've biological replicates from different samples interspersed. This would mean that the biological variability of replicates is higher than the biological variability that you're trying to measure across a condition/treatment/gradient. In my opinion a PCoA is a good statistical test to assess the variability of a community and biological replicates of a community.

I hope this helps.

António