enrichment

Enrichment & Functional Annotation

This skill guides method selection and execution for gene set enrichment, over-representation analysis, per-sample pathway scoring, and transcription factor activity inference.

Method Selection Decision Tree

Choose the method based on your input data and analytical goal:

1. Input is a ranked gene list (with scores, fold-changes, or test statistics)

   • Use GSEA via gseapy.prerank() (Python) or fgsea::fgsea() (R via rpy2).
   • Rank by signed statistic (e.g., sign(log2FC) * -log10(pvalue)), not by p-value alone.

2. Input is an unranked gene list (e.g., DE genes at FDR < 0.05)

   • Use ORA via gseapy.enrich() (Python) or clusterProfiler::enrichGO() (R via rpy2). Do NOT use enrichKEGG() — KEGG is not staged and the call needs network access it will not get.
   • Pass gene sets as a file resolved from the reference data available to you, never as an Enrichr library name string — those trigger HTTP requests and fail.
   • ALWAYS supply the background gene set (all expressed/detected genes).

3. Need per-sample pathway activity scores

   • Use ssGSEA via gseapy.ssgsea() for MSigDB-style gene sets.
   • Use GSVA via gsva() (R via rpy2) when downstream analysis expects continuous scores per pathway per sample.

4. Pathway activity on AnnData (single-cell or bulk)

   • Use decoupler dc.mt.ulm() or dc.mt.mlm() with the PROGENy model.
   • Results integrate directly into adata.obsm (score_ulm / padj_ulm) for downstream plotting and clustering.

5. Transcription factor activity inference

   • Use decoupler with CollecTRI regulon resource (NOT enrichment databases).
   • This is TF activity, not pathway enrichment -- use the correct framing.

Gene Set Database Selection

Ask for a database by name and resolve it from the reference data available to you — never assume a path, a filename, or a format. What is provisioned varies per environment, so confirm what you have before committing to a method.