omicverse.metabol.asca

Contents

omicverse.metabol.asca#

omicverse.metabol.asca(adata, *, factors, include_interactions=True, n_components=2, n_permutations=0, layer=None, center=True, seed=0)[source]#

ASCA — ANOVA-Simultaneous Component Analysis (Smilde 2005).

Decomposes the mean-centered data matrix into per-factor effect matrices plus (optionally) pairwise interactions and a residual, runs PCA on each effect, and reports variance explained and (optionally) a permutation p-value.

Parameters:

  • factors (Sequence[str]) – 1–3 column names in adata.obs. Each is treated categorically.

  • include_interactions (bool (default: True)) – Whether to add pairwise A:B interaction terms. Three-way interactions are not computed here — in practice they require far more replicates than a typical metabolomics design supports.

  • n_components (int (default: 2)) – Number of PCs to retain per effect. Default 2 — enough for the standard ASCA scatter plot.

  • n_permutations (int (default: 0)) – If > 0, shuffle each factor’s labels that many times and compute the null distribution of effect SS. Reported as p_value. Default 0 (fast; no p-value).

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

  • center (bool (default: True)) – If True (default) subtract the grand mean before decomposition. ASCA as described in Smilde 2005 assumes centered data.

  • seed (int (default: 0)) – RNG seed for permutation.

Returns:

Access effects via result.effects[name] or call ASCAResult.summary(). Names are factor names (e.g. "treatment") or "A:B" for interactions.

Return type:

ASCAResult

Contents