omicverse.micro.meta_da

Contents

omicverse.micro.meta_da#

omicverse.micro.meta_da(studies, group_key, group_a=None, group_b=None, method='deseq2', rank='genus', min_prevalence=0.1, combine='random_effects', study_names=None, **method_kwargs)[source]#

Per-study DA + inverse-variance meta-analysis.

Parameters:

  • studies (Sequence[AnnData]) – List of per-study AnnData objects.

  • group_key (str) – Same semantics as ov.micro.DA.wilcoxon() / DA.deseq2() / DA.ancombc(). If group_a / group_b are omitted, the two sorted unique values of group_key in the first study are used (and re-used for every study).

  • group_a (Optional[str] (default: None)) – Same semantics as ov.micro.DA.wilcoxon() / DA.deseq2() / DA.ancombc(). If group_a / group_b are omitted, the two sorted unique values of group_key in the first study are used (and re-used for every study).

  • group_b (Optional[str] (default: None)) – Same semantics as ov.micro.DA.wilcoxon() / DA.deseq2() / DA.ancombc(). If group_a / group_b are omitted, the two sorted unique values of group_key in the first study are used (and re-used for every study).

  • method (str (default: 'deseq2')) – 'wilcoxon', 'deseq2', or 'ancombc'. The per-study effect sizes must be on a log-fold-change scale; Wilcoxon is supported but its reported log2FC has no standard-error, so Wilcoxon meta-DA uses the empirical between-study SE to weight (i.e. every study gets unit weight pre-τ² — still useful as a sanity check).

  • rank (Optional[str] (default: 'genus')) – Collapse to this taxonomic rank in every study before DA, so features align across cohorts. None assumes the studies already share the same feature ids.

  • min_prevalence (float (default: 0.1)) – Passed through to each per-study DA call.

  • combine (str (default: 'random_effects')) – 'random_effects' (default; DerSimonian-Laird τ²) or 'fixed_effects'.

  • study_names (Optional[Sequence[str]] (default: None)) – Labels for the per-study result columns; defaults to ['study_0', 'study_1', …].

  • **method_kwargs – Extra kwargs forwarded to the underlying DA call (e.g. pseudocount=0.5 for ancombc).

Return type:

DataFrame

Returns:

  • DataFrame indexed by feature with columns

    • combined_lfc — meta-analytic log2 fold-change estimate

    • combined_se — standard error of the combined estimate

    • z — Wald z-score (combined_lfc / combined_se)

    • p_value / fdr_bh — two-sided p + BH-FDR

    • n_studies — number of cohorts in which the feature was tested

    • Q — Cochran’s Q statistic of between-study heterogeneity

    • I2 — I² heterogeneity (0 → homogeneous, > 75% → high)

    • tau2 — between-study variance (random-effects only)

    • per-study columns lfc_<study> and se_<study> for traceability

Contents