Shannon diversity: not normally distributed

Downstream analysis and statistics MaAsLin
1

Hello,

I am using MaAsLin2 to examine the relationship between a categorical variable of mood (happy/neutral/not happy) and Shannon diversity. My Shannon diversity does not have any 0’s. The Shannon diversity range is 0.3 to 3.5, and the average is 2.2.

I am using the following code, adjusting for covariates:

fit_data = Maaslin2(
input_data = Shannon_diversity,
input_metadata = metadata,
output = “Adjusted Shannon Results”,
analysis_method = “LM”,
normalization = “CLR”,
transform = “NONE”,
standardize = TRUE,
min_prevalence = 0,
min_abundance = 0,
plot_heatmap = TRUE,
fixed_effects = c(“mood”, “Infant_ethnicity”, “infant_sex”),
reference = c(“mood,happy”, “Infant_ethnicity,Caucasian”, “infant_sex,Male”))

My issue is as follows: Shannon diversity is not normally distributed, so I am using CLR normalization. This does give me significant findings for mood, however the coefficients are exponentials (e.g. -5.9 e-17 ± 1.7 e-17, p = 0.0005, q = 0.01).

Can someone please advise why this may be, and whether this is correct or erroneous?

2

Hi,

MaAsLin 2 is designed to analyze the association between metadata (like mood) and taxonomic abundances (the actual relative abundance of each taxon), not summary statistics like Shannon diversity. Presumably you have one Shannon diversity value per sample and one set of metadata per sample. In that case case, you probably want to run a standard linear model like:

lm(shannon ~ mood + Infant_ethnicity + infant_sex)

You might also decide to transform shannon beforehand or use robust linear modeling if you know it is not normally distributed.

Will