Is MaAsLin2 reading features all at once?

Downstream analysis and statistics MaAsLin
1

Hello, I know that MaAsLin2 does not consider all features (dependent variables such as taxa) together in a single analysis. However, I am encountering an unusual problem:

  1. I have one metadata file and two data files (data file 1 and data file 2).

  2. The two data files have a number of dependent variables (taxa), some of which are unique to the data file 1 and data file 2, and a couple are the same dependent variable that is just present in both data file 1 and data file 2 (e.g., OTU_Shannon).

  3. When I use the same metadata against two different data files, I get different responses for associations between my metadata and the identical shared dependent variable (e.g., OTU_Shannon).

My codes are below - does someone know what the issue may be here?

fit_data = Maaslin2(
input_data = data_file_1,
input_metadata = metadata,
output = “Energy”,
analysis_method = “LM”,
normalization = “NONE”,
standardize = TRUE,
min_prevalence = 0.1,
min_abundance = 0.001,
plot_heatmap = TRUE,
fixed_effects = c(“energy”))

fit_data = Maaslin2(
input_data = data_file_2,
input_metadata = metadata,
output = “Energy”,
analysis_method = “LM”,
normalization = “NONE”,
standardize = TRUE,
min_prevalence = 0.1,
min_abundance = 0.001,
plot_heatmap = TRUE,
fixed_effects = c(“energy”))

Again, the metadata is identical: “energy” is going up against data file 1 and data file 2, both of which have the same dependent variable (OTU_Shannon) but also unique dependent variables (specific taxa). In data file 1, energy is associated with OTU_Shannon, but when using data file 2 it is not.

2

Here’s another example looking at an amino acid from the metadata: again, the metadata used in association with data file 1 and data file 2 is identical, and OTU_Observed is also identical between data file 1 and data file 2. Below, you will see I get different FDR’s even though you can see I am working with the exact same data points and exact same code. I also have a data file 3 with the same OTU_Observed and that one doesn’t come up significant at all. Could the other taxa being different in each of these datasets be affecting the results?

gln_1
gln_1960Ă—960 46.4 KB

gln_2
gln_2960Ă—960 46.4 KB

@andrewGhazi @himel.mallick @nearinj I would appreciate your help.

3

Update 3: I created test datasets mimicking the original data I was working with:

Metadata: energy
Dataset 1: OTU_Shannon, bifido, lacto
Dataset 2: OTU_Shannon, proteo, bacteroides, clostridium

OTU_Shannon are identical to one another.

When running MaAslin2, energy is associated with OTU_Shannon in dataset 1 but not in dataset 2. Can someone please explain why that is? I was under the impression that MaAslin2 examines the relationship between a specific metadata (energy) and each individual feature separately.

Thank you.

4

hi there,

It looks like the results are as expected given the coefficients are the same. The key thing to remember is that the q-value output is an FDR corrected p-value (False discovery rate - Wikipedia). This means that the number of tests and the distribution of those p-values in each Maaslin run will impact the final q-values for each variable (even if they have the same coefficient and same p-values.

Hope that helps
Cheers,
Jacob Nearing

1 Like