omicverse.metabol.dgca

Contents

omicverse.metabol.dgca#

omicverse.metabol.dgca(adata, *, group_col, group_a=None, group_b=None, features=None, method='pearson', abs_r_threshold=0.3, layer=None)[source]#

Differential correlation between two groups.

Parameters:

  • group_col (str) – Column in adata.obs with the two-class labels.

  • group_a (Optional[str] (default: None)) – Labels to contrast. None → first two unique values. The z-diff direction is z_a - z_b (so positive z_diff means correlation is stronger in group A).

  • group_b (Optional[str] (default: None)) – Labels to contrast. None → first two unique values. The z-diff direction is z_a - z_b (so positive z_diff means correlation is stronger in group A).

  • features (Optional[Sequence[str]] (default: None)) – Optional list of metabolite names to restrict the pair enumeration to. None → all. For p features the output has p*(p-1)/2 rows.

  • method (str (default: 'pearson')) – "pearson" (default) or "spearman". Spearman is rank-based and more appropriate for non-Gaussian metabolite distributions; Pearson is faster and matches the DGCA default.

  • abs_r_threshold (float (default: 0.3)) – Threshold for calling a correlation “significant in group X” when assigning the dc_class label. Default 0.3 matches MetaboAnalyst’s network module.

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

Returns:

Long format, one row per feature pair, columns:

  • feature_a, feature_b — metabolite names

  • r_a, r_b — correlation in each group

  • z_diff — Fisher-z-transformed difference test statistic

  • pvalue, padj — two-sided p and BH-FDR

  • dc_class — "+/+", "+/0", "+/-", "0/+", "0/-", "-/+", "-/-", or "0/0". The first symbol is group A, the second group B. + / - means |r| ≥ abs_r_threshold with that sign; 0 means below threshold.

The result is sorted by padj ascending.

Return type:

pd.DataFrame

Contents