TSS normalization of normalized metabolomics data

I read that the authors of MaAsLin do not support analysis of metabolomics data with MaAsLin2.

However, I know that many people have been experimenting with this.

I have tried several things. I originally normalize the metabolomics data using the conventional mean centering and log transformation and a density plot of such normalized data looks nice. However, pretty much nothing is significant after MaAsLin fitting with the parameter normalization=‘NONE’ When I run MaAsLin with the same normalized data, but with default MaAsLin2 normalization (TSS), few features (metabolites). appear significant A density plot of normalized metabolomics data plus TSS, also appears normal but compressed around zero (a way narrower and taller bell curve; almost like a peak). QUESTION: Is it incorrect to apply TSS to data that has already been normalized by mean entering and log transformation, as is typical for metabolomics data? THANKS!

Hi @jjovel

We had a fairly long discussion around this and we came to the conclusion that while TSS normalization might be defensible we would normally suggest using median normalization as sometimes metabolites that have really high abundances (such as water) can result in difficult to interpret TSS values.

Hopefully this is helpful

Jacob Nearing