Transformations and results interpretations


I have been playing around with different transformations with Maaslin2. I have relative abundance data, so AST transformation would fit my data, being between 0-1. I was not sure whether the results from Maaslin2 are still in the transformed scale, so I ran a linear model outside of Maaslin2 after transforming my taxa to AST and could replicate the same results from Maaslin2, meaning that the results from Maaslin2 are not back-transformed, which I have to keep in mind for interpretation.

Now, when I look at the log-transformation on Maaslin2. I ran the linear models outside of Maaslin2 and compared, but they are not consistent neither with natural-log or with log10, nor with their back-transformations.

My questions are the following:

  1. Does Maaslin2 use natural logarithm or log10 transformation?
  2. Does it give results in the logarithmized scale, or does it back-transform?

I am very surprised that my attempt to replicate the results worked for AST but not for LOG. Your answers will be very helpful in trying to understand what is going on and how I should interpret my results.

Thanks so much!


To answer your first question, it uses log2(). I believe the results are not transformed back to the original scale.

Thanks for your quick reply. It all makes sense now!