Interpretation of interaction terms

Downstream analysis and statistics MaAsLin
1

Hi,

I confused myself about the interpretation of interactions terms in Maaslin3.

Here is my model, where Group can either be Control or Probiotics, and Responder can either be R or NR:

maaslin3(input_data = input_maaslin,
input_metadata = metadata_maaslin,
output = “m3_output”,
formula = ‘~ Group * Responder + (1|Individual)’,
normalization = ‘TSS’,
transform = ‘LOG’,
augment = TRUE,
standardize = TRUE,
max_significance = 0.25,
median_comparison_abundance = TRUE,
median_comparison_prevalence = FALSE)

In the usual table output, i get significant results for the term GroupProbiotics:ResponderR, and I am unsure on how to interpret it.
Let’s say the abundance coef is -1 for a taxa, does that mean that in the Probiotics group, I have 2 times less of that taxa in the R compared to the NR?
Basically that I’m looking at the R/NRdifference for the “fixed” Probiotics group?

This is what I understood from the dedicated section of the tutorial, but I got confused after looking at other posts on the forum.

Thanks a ton!

2

Hi,

I’ll assume your base level groups are Control and NR since those are alphabetically first, and the interaction term name you get (GroupProbiotics:ResponderR) should be for the non-baseline groups. Your model should have 3 terms: GroupProbiotics, ResponderR, and GroupProbiotics:ResponderR. Because the model is fully interacted, GroupProbiotics is the log2 fold change of the probiotic group relative to the control group among the non-responders; ResponderR is the log2 fold change of the responders relative to the non-responders in the control group; and GroupProbiotics:ResponderR is the additional fold change of the probiotic group relative to the control group among the responders (i.e. how different this is from the effect in the non-responders). For example, if GroupProbiotics:ResponderR was 0, there would be no difference in the effect of the probiotic for responder vs. nonresponder. However, there could still be an effect of probiotic vs. control within both groups - that’s what the GroupProbiotics tells you - it just has to be the same effect.

To give a clear answer to your original question, this is not a correct statement:

Let’s say the abundance coef is -1 for a taxa, does that mean that in the Probiotics group, I have 2 times less of that taxa in the R compared to the NR?

For comparing R to NR in the Probiotics group, your log2 difference is ResponderR + GroupProbiotics:ResponderR, not just GroupProbiotics:ResponderR. ResponderR tells you the difference for R vs NR in the controls, but GroupProbiotics:ResponderR is the additional difference for the probiotic group, not the direct difference.

Let me know if that’s still confusing.

Will

3

Thank you so much for this detailed and very clear answer, it’s super helpful!
I’m glad I asked because I definitively didn’t had it right.