Technical vs Biological replicates

[lsulli10]

I really like this package overall - it's been quite straightforward to use. I've been trying to find in the documentation if there is a way to differentiate between biological and technical replicates for analysis - I would like to be able to use all the technical replicates for imputation in particular, but only if I can do so without incorrectly altering my sample size.

Thanks very much!

[wolski]

Hallo @lsulli10,

Do I understand correctly that you have a design that looks like

Group: A, A, A, A,
Biorep : A, A, B, B,
Techrep: 1, 2, 1, 2

?

So, in this case, two technical replicates per biological replicate.

Since the technical replicates are highly correlated and not IID observations, you can not use them for inference, but you would still like to use all of them instead of selecting only one (either 1 or 2).

The easiest way to address it is to aggregate the technical replicates. You could do it in prolfqua using the lmrob (robust linear model) or medpolish.

To do it, you include the technical replicate in your hierarchy description. The hierarchy is used to describe the hierarchy of your measurement in proteomics, typically ion -> peptidoform -> stripped sequence -> protein
You could also include the technical replicate here.
Starting from here:
https://fgcz.github.io/prolfqua/articles/CreatingConfigurations.html

You can extend it by the technical replicate:

atable$hierarchy[["proteinID"]]    <-  "proteinID"
atable$hierarchy[["techrep"]] <- "Techrep"
atable$hierarchy[["peptideID"]]    <-  "peptideID"

If you start from protein level data your hierarchy would look like this:

atable$hierarchy[["proteinID"]]    <-  "proteinID"
atable$hierarchy[["techrep"]] <- "Techrep"

Now use the LFQDataAggregator class to roll up the data to protein level:

https://fgcz.github.io/prolfqua/reference/LFQDataAggregator.html

Since, specifically, the medpolish method robustly infers protein abundances, in the presence of missing observations, you will have fewer missing observations than when using one of the technical replicates.

If you provided more details about your experiment, I could give you a better, more tailored reply.

Witold