omicverse.metabol.mixed_model

Contents

omicverse.metabol.mixed_model#

omicverse.metabol.mixed_model(adata, *, formula, groups, re_formula='1', term=None, layer=None)[source]#

Per-feature statsmodels.MixedLM fit.

Fits y ~ <formula> for each metabolite with groups=<groups> defining the random-effect grouping variable (e.g. patient ID).

Parameters:

  • formula (str) – Patsy-style formula without the y ~ prefix. Categorical effects, interactions (a * b), and numerics all work.

  • groups (str) – Name of the adata.obs column carrying the random-effect grouping labels. Usually patient / subject ID.

  • re_formula (Optional[str] (default: '1')) – Random-effect formula. "1" (default) = random intercept. Use "1 + time" for random slopes.

  • term (Optional[str] (default: None)) – If given, return a short-format table with one row per feature for this specific fixed-effect term (matches the metabol.differential() schema: stat / pvalue / padj). If None, return a long-format table with one row per (feature, term) pair, excluding the intercept and variance components.

  • layer (Optional[str] (default: None)) – AnnData layer name (default None → adata.X).

Returns:

Long format (default): columns feature, term, coef, se, stat, pvalue, padj. Short format (term=...): indexed by feature with coef, se, stat, pvalue, padj.

Return type:

pd.DataFrame

Contents