/gseapy-gene-enrichment

name: "gseapy-gene-enrichment"
description: "GSEA and over-representation analysis (ORA) for RNA-seq and proteomics. Wraps Enrichr for ORA against MSigDB, KEGG, GO, and 200+ databases; runs preranked GSEA on ranked DE gene lists. Outputs enrichment tables and running-score plots. Use after DESeq2 or edgeR for pathway-level interpretation."
license: "MIT"

GSEApy — Gene Set Enrichment Analysis in Python

Overview

GSEApy provides Python implementations of GSEA and over-representation analysis (ORA) for interpreting gene expression changes at the pathway level. The `enrich` module queries the Enrichr API to test a gene list against 200+ databases (GO, KEGG, MSigDB Hallmarks, Reactome, WikiPathways). The `prerank` and `gsea` modules run the GSEA algorithm on a pre-ranked gene list or expression matrix — computing normalized enrichment scores (NES) and FDR values for each gene set. GSEApy integrates directly with pandas DataFrames from DESeq2 or scanpy differential expression output, making it the standard Python tool for pathway analysis in RNA-seq workflows.

When to Use

• Interpreting DESeq2 or edgeR differential expression results at pathway/GO-term level
• Running fast ORA (over-representation analysis) against Enrichr's 200+ databases including GO, KEGG, and MSigDB Hallmarks
• Performing GSEA prerank analysis on a log2-fold-change-ranked gene list without an expression matrix
• Identifying enriched pathways in scRNA-seq cluster marker genes
• Generating publication-ready enrichment dot plots and GSEA running-score plots
• Use **GSEA Java application** for the official GUI-based analysis with full GSEA desktop interface
• Use **fgsea** (R) as an alternative with fast permutation-based p-values; GSEApy is preferred for Python-native pipelines
• Use **omics-plotting** SKILL after DE and for publication-quality plots of gsea results

Prerequisites

• **Python packages**: `gseapy`, `pandas`, `matplotlib`
• **Internet access**: `enrich` module queries the Enrichr API (requires connection)

pip install gseapy

# Verify
python -c "import gseapy; print(gseapy.__version__)"
# 1.1.3

Pre-flight Interview

Settle these with the user before writing any analysis code.

decisions:
  - id: D1
    param: analysis_mode
    kind: required
    source: upstream
    ask: "Test only the genes that passed a significance cut (ORA), or the whole ranked list without a cut (GSEA prerank)?"
    default: "prerank when the upstream stage yields a full ranking; ORA when it yields a gene list"

  - id: D2
    param: organism
    kind: derived
    source: upstream
    ask: "Which species are these genes from?"
    default: "carried from the upstream dataset"

  - id: D3
    param: gene_sets
    kind: required
    source: user
    depends_on: [D2]
    ask: "Which collections should the genes be tested against - GO terms, pathways, disease or perturbation signatures?"
    default: null

  - id: D4
    param: rnk
    kind: optional
    source: user
    depends_on: [D1]
    ask: "What should order the genes - effect size, signed significance, or the test statistic?"
    default: "log2FoldChange"
    skip_if: "ORA mode - ranking is not used"

  - id: D5
    param: cutoff
    kind: optional
    source: user
    ask: "How strong must a term's evidence be before it is reported?"
    default: 0.05

  - id: D6
    param: min_size, max_size
    kind: optional_conditional
    source: user
    ask: "Should very small or very broad gene sets be excluded as uninformative?"
    default: "15-500 genes"

  - id: D7
    param: permutation_num
    kind: optional_conditional
    source: user
    ask: "More permutations give finer p-values at proportionally more runtime - is the default enough?"
    default: 1000

  - id: D8
    param: weighted_score_type
    kind: optional
    source: user
    ask: "Should GSEA use the standard weighted enrichment score, or a different weighting for this ranking?"
    default: 1

  - id: D9
    param: seed, threads
    kind: never_ask
    source: data
    reason: "A fixed seed and available CPU cores affect reproducibility and runtime, not the enrichment model."
    default: "seed fixed, threads min(4, available_cores)"

Two orderings matter here. D3 hangs on D2 because Enrichr's libraries are species-specific — the catalog itself changes with the organism. D4 hangs on D1 because a ranking metric only exists in prerank mode; under ORA the question has no answer, which is what `skip_if` records.

Quick Start

import gseapy as gp

# ORA: test a gene list against GO Biological Process
gene_list = ["TP53", "BRCA1", "CDK2", "CCND1", "MYC", "EGFR", "KRAS", "PTEN"]

enr = gp.enrichr(gene_list=gene_list,
                 gene_sets=["GO_Biological_Process_2023"],
                 organism="human",
                 outdir=None)
print(enr.results.head(5)[["Term", "P-value", "Adjusted P-value", "Genes"]])

Workflow

Step 1: Over-Representation Analysis with Enrichr (ORA)

Test a gene list against pathway databases via the Enrichr API.

import gseapy as gp
import pandas as pd

# Gene list from DESeq2 (significant upregulated genes)
sig_genes = ["TP53", "BRCA1", "CDK2", "CCND1", "MYC", "EGFR",
             "KRAS", "PTEN", "RB1", "AKT1", "PIK3CA", "MDM2"]

# Run ORA against multiple databases
enr = gp.enrichr(
    gene_list=sig_genes,
    gene_sets=[
        "GO_Biological_Process_2023",
        "KEGG_2021_Human",
        "MSigDB_Hallmark_2020",
        "Reactome_2022",
    ],
    organism="human",
    outdir="enrichr_results/",
    cutoff=0.05,
)

# Display top results
results = enr.results
print(f"Enriched terms: {len(results[results['Adjusted P-value'] < 0.05])}")
print(results[results["Adjusted P-value"] < 0.05].sort_values("Adjusted P-value")
      .head(10)[["Gene_set", "Term", "Adjusted P-value", "Combined Score"]])

Step 2: List Available Gene Set Databases

Discover the 200+ databases available through Enrichr.

im