MaAsLin2

Coming soon

The MaAsLin2 runner isn't wired yet. The module slot is reserved so the Microbiome portal's left-hand nav stays stable once the runner ships — analysis records can still be created and saved through the admin in the meantime.

What it will do

MaAsLin2 (Multivariate Association with Linear Models) fits per-feature generalised linear (or linear-mixed) models between microbial features — taxa, pathways, functions — and sample-level metadata.

Parameters the module already captures on its model, ready for the runner:

• fixed_effects (list): covariates treated as fixed (e.g. Study.Group, Age).
• random_effects (list): grouping variables (e.g. Subject, Site).
• reference_levels (dict): categorical baselines (e.g. Study.Group → Control).
• normalization: TSS, CLR, CSS, TMM, NONE. Default TSS.
• transform: LOG, LOGIT, AST, NONE. Default LOG.

Comparator positioning

MaAsLin2 is the natural comparator to GeMMA's punch-table CLR+Wilcoxon row because it asks the same question (which microbial features shift with metadata?) using a richer modelling framework: multiple covariates at once, correct for random effects, and handle compositional data explicitly.

Runner bring-up plan

1. Add an R-bridge container (R + MaAsLin2 package) to docker-compose.yml.
2. Extend MaaslinAnalysis with fixed/random-effect wiring from the UI.
3. Ship a MaAsLin2 results template analogous to humann3.html.
4. Point the CLI comparator (benchmark_maaslin) at the same cohort as benchmark_gemma so the paper's main figure can stack all three results side-by-side.