Normalization of a metabolomics dataset: is it too much?

Question

I am working with a dataset of quantitative targeted LC-MS/MS-measured metabolites. I am advised to perform following normalization steps:

1. Removal of metabolites where >20% of samples are below LOD (there are no missing values)
2. Log-transformation
3. Merging of both batches via ComBat
4. Quantile normalization

Moreover for multivariate analysis I should use pareto scaling. From my understanding all these steps (except 1.) are some kinds of normalization methods. So my questions are:

1. Isn't this too much and/or partly redundant?
2. If it is not per se too much, does it make sense in the first place?
3. Does the suggested order of these steps make sense?

I do know all these methods and that individually they make sense. However, I couldn't find much about the combination of these. I also tried to use the NOREVA tool but I couldn't get it to work while I tried many different browsers on different OSs. The code for NOREVA (which is a collection of tests to evaluate the effect fo different normalization strategies) is not available and it would probably take me a long time to implement all these tools by myself. So I first wanted to ensure if anyone has any insights on the proposed normalization strategy before choosing some criteria to evaluate it.

Thanks in advance!

Answer 1

Noreva works fine for me. Might be a temporary issue. With Mozilla and Chrome both.

Well, with Noreva you can include "QCs" from different (2 batches, right?) to correct for batch effects. Corrections for batches without QC options also available on Noreva.

1. No steps are very logical. 1/5th of LOD values are a general cut-off to use- nothing wrong there.

2. Yes, they all make sense.

3.To me, median normalization worked better than Quantile, and cube root transformation worked better than "log2" transformation- but really can be platform specific and the number of variables you have !

Thanks,