Hi there,
Could you please advise the difference in how you apply the pseudocount in Maaslin2 vs Maaslin3 when using LOG transform or CLR? I see that in Maaslin3 you split out the non-zero values, but how does this impact features which have both zero and non-zero values? I am a little confused about this.
Many thanks,
Katherine
In MaAsLin 2, we essentially make the assumption that all zeros correspond to taxa that are present but below the limit of detection. In MaAsLin 3, we assume that the zeros correspond to true absences. Practically, in MaAsLin 3, we use a linear model for the nonzeros after log transformation, and we use a logistic model for zeros vs. nonzeros. In particular, a pseudocount of half the minimum abundance is used in MaAsLin 2, but no pseudocount is used in MaAsLin 3. Let me know if that answers your question.
Will
It does, thanks Will.