Hi, I am trying to test if my samples microbiome is predictive of my infection severity
I have the “pre-infection microbiome” of my samples and I have the pathogen load over course of infection or “infect integral”. I am wondering what variables I should put into Maaslin.
In MaAsLin2, can I place “infection integral” as fixed effects, or would that imply it was causing the community difference?
Hi @Rachael_Kramp - if I understand correctly, you have a variable (or a set of variables) named infection severity (let’s call it Y) and pre-infection microbiome samples (X) and you want to test whether X is predictive of Y? Could you share more details such as how many Y features and sample size? Depending on the question and data dimensions, MaAsLin 2 may or may not be appropriate for this task.
You are correct I am trying to see if microbial community before infection is correlated with severity. We used 16s metabarcoding and have around 40 microbiome samples (X) and their corresponding (Y) values. The infection severity is a continuous value generally between 0-1200.
OK - in that case, you can use infection severity as a fixed effect to identify individual features statistically significant with respect to infection severity. Note that, if you are interested in community-level inference instead, PERMANOVA is a better approach for that purpose. Hope this helps!
Hi Himel,
I just read this post this post where you discussed how MaAsLin2 is appropriate to find if 16s compositional abundance features are significantly correlated with a continuous variable, but you suggested that PERMANOVA is better for “community level inference”. What do you mean by “community level”? I have 16s data grouped by each level taxonomy from phylum to species, and I want to make sure you’re not saying that only particular “taxonomy levels” are appropriate to use with this kind of analysis in MaAsLin2. Thanks!
Hi there,
What I believe Himel is referring to with the “community level” inference is if a metadatum feature explains an overall change in the composition of the microbiome, instead of diving into the per-feature level differences. The overall composition level differences can be investigated with the PERMANOVA type (e.g. adonis of similar) analysis.
As for the taxonomy level- I would only use either one tax level at a time or use terminal taxa, but any of the levels can be used through MaAsLin. Basically make sure no OTUs are represented more than once while running the model.
I hope that helps!
Best,
Kelsey