Controlling for nested matched controls

Downstream analysis and statistics MaAsLin
1

Hello,

I’m working on doing some DA testing between genefamily abundance tables of IBD and healthy control patients. The study is designed with nested cohorts, where patients are matched based on very similar covariates. For example, in 100 patient samples, there would be 20 nested cohorts of 5 patients, each with 1 IBD and 4 healthy matched controls.

I’ve given each of these cohorts a unique cluster identifier which I use in the random effects argument when calling maaslin2 as such;

results <- Maaslin2(
  input_data = filtered_abundance,
  input_metadata = filtered_meta,
  output = "/Volumes/PGH-Backup/x/maaslin2",
  fixed_effects = c("diagnosis"),
  random_effects = c("cluster"),
  normalization = "none", 
  transform = "none",
  reference = c("diagnosis,healthy")
)

Im wondering if this would be an appropriate way to do the analysis? Could issues arise if there is 1 sample in a cohort, or a cohort with no matched IBD patient?

Thank you!

2

Hi @immunochem.

MaAslin2 does not currently support nested random effects. My best suggestion would be to use the lowest level of the nested effects as your random effect.

Cheers,
Jacob Nearing