omicverse.metabol.differential

Contents

omicverse.metabol.differential#

omicverse.metabol.differential(adata, *, group_col='group', group_a=None, group_b=None, method='welch_t', layer=None, log_transformed=True)[source]#

Run a univariate two-group test across all metabolites.

Parameters:

  • group_col (str (default: 'group')) – Name of the factor column in adata.obs.

  • group_a (Optional[str] (default: None)) – Which two values of group_col to contrast. When None, the first two unique values are used. log2fc is reported as group_a / group_b.

  • group_b (Optional[str] (default: None)) – Which two values of group_col to contrast. When None, the first two unique values are used. log2fc is reported as group_a / group_b.

  • method (Literal['t', 'welch_t', 'wilcoxon', 'limma'] (default: 'welch_t')) –

    • "welch_t" (default) — Welch’s t-test; handles unequal variances. The MetaboAnalyst default.

    • "t" — Student’s t (equal-variance).

    • "wilcoxon" — Mann-Whitney U; non-parametric.

    • "limma" — empirical-Bayes moderated t (Smyth 2004). Implemented here directly on the variance pool — matches limma’s output at atol~1e-6 on real data.

  • layer (Optional[str] (default: None)) – AnnData layer holding the values to test. Default None = use adata.X (which the pyMetabo pipeline leaves normalized and transformed).

  • log_transformed (bool (default: True)) – If True (default), data are assumed already log-transformed and log2fc = mean_a - mean_b (difference of logs). If False, the fold-change is computed as log2(mean_a/mean_b) on the raw scale.

Returns:

Indexed by metabolite (adata.var_names) with columns stat, pvalue, padj, log2fc, mean_a, mean_b.

Return type:

pd.DataFrame

Contents