omicverse.micro.DA.deseq2

Contents

omicverse.micro.DA.deseq2#

DA.deseq2(group_key, group_a=None, group_b=None, rank=None, min_prevalence=0.1, alpha=0.05)[source]#

Differential abundance via pyDESeq2 (negative-binomial GLM).

Models raw integer counts directly with a NB GLM — no prior CLR / log transform. Mature method from RNA-seq adapted to microbiome composition; tends to be more conservative than Wilcoxon and has explicit shrinkage of small-count log-fold-changes. ANCOM-BC is closer to the compositional ground-truth but pyDESeq2 is faster and more widely cited.

Parameters:

  • group_key (str) – Column in adata.obs with the binary phenotype.

  • group_a (str, optional) – Class labels; if omitted, the two alphabetically smallest values in adata.obs[group_key] are used. Convention: log2fc > 0 ⇒ feature higher in group_a.

  • group_b (str, optional) – Class labels; if omitted, the two alphabetically smallest values in adata.obs[group_key] are used. Convention: log2fc > 0 ⇒ feature higher in group_a.

  • rank (str, optional) – Collapse counts to this taxonomic rank ('genus', 'family', …) before testing; None uses ASVs.

  • min_prevalence (float, default 0.1) – Drop features present in fewer than this fraction of samples before testing.

  • alpha (float, default 0.05) – FDR threshold passed to pyDESeq2’s summary.

Returns:

  • pd.DataFrame indexed by feature with columns baseMean,

  • log2fc, lfcSE, stat, pvalue, padj.

Contents