henhaobio
Sign in

Nature ProtocolsBioinformaticsPeer-reviewed journal article

Pathway enrichment analysis of -omics data

Jüri ReimandiD, Ruth IsserliniD, Véronique VoisiniD, Mike Kucera, Christian Tannus-Lopes, Asha Rostamianfar, Lina WadiiD, Mona Meyer, J. D. Wong, Changjiang Xu, Daniele MericoiD, Gary D. BaderiD

PDFDOIPreprintCodeData

In-depth summary

Claude Opus 5.5 · xhigh effort · 7 Nov 2026 · from the full text of version 1 (12 Dec 2017) · CC BY 4.0

Step-by-step protocol for interpreting omics gene lists via g:Profiler or GSEA enrichment testing, then grouping redundant enriched pathways into themes with Cytoscape Enrichment Map.

Omics experiments produce long gene lists that would take an impractical amount of manual literature searching to interpret, and many biologists have no bioinformatics training. The protocol turns such lists into interpretable pathway themes with free point-and-click tools, and it advises on test choice, gene set size limits, background sets and multiple-testing thresholds.

Key results

  1. ShownThe gene-list route is shown by running g:Profiler's ordered Fisher's exact test on cancer driver genes from The Cancer Genome Atlas (TCGA) against GO biological processes and Reactome. The output is exported in Generic Enrichment Map format.

    Input: 127 MuSiC-defined driver genes from TCGA exome data of 3,200 tumors of 12 types. Settings: ordered query, GO terms inferred from electronic annotation (IEA) excluded, pathway size 5 to 350, at least 3 query/term overlaps, corrected q < 0.05 reported. Runtime about 3 minutes in Chrome on Windows7. · INTRODUCTION; Equipment; Part 1A - Pathway enrichment analysis of a gene…; TIMING

  2. ShownThe ranked-list route is shown with GSEAPreranked (gene set permutation), contrasting the mesenchymal and immunoreactive TCGA ovarian cancer subtypes. It uses a GMT file compiled from seven pathway sources, with IEA GO terms excluded.

    Default 1000 permutations; maximum set size lowered from 500 to 200; about 20 minutes on Windows7 with 1GB RAM and Java 8. On a 16G RAM, i7 3.4 GHz machine, 10, 100, 500 and 1000 randomizations took 155, 224, 544 and 1012 seconds. · Equipment; Part 1B - Pathway enrichment analysis of a…; TIMING; TROUBLESHOOTING

  3. ShownEnrichment Map in Cytoscape draws enriched pathways as a gene-overlap network. AutoAnnotate (using clusterMaker2 and WordCloud) then groups and labels redundant pathways as themes for collapsed, sub-network and publication figures.

    Maps built at FDR q < 0.01 with combined coefficient > 0.375 (combined constant 0.5). NES filters at 2.5 and -2.5 selected 32 nodes for a sub-network. TP53 and BGN are the top genes in the g:Profiler and GSEA results. About 5 minutes to build and about 4 hours to annotate, with 8GB RAM. · Protocol 2 - Visualize enrichment results with…; TIMING; Protocol 3 - Navigating and interpreting the…; TIMING

  4. ClaimedThe test should match the list type: Fisher's exact test for unranked lists (with a custom background for targeted panels or phosphoproteomics) and GSEA for genome-wide ranks. Gene set permutation suits few replicates; phenotype permutation suits more.

    Gene set permutation is advised for 2 to 5 replicates per condition, phenotype permutation above 6 to 10. Exclude pathways under ten or fifteen genes and, for human expression data, over 300 genes (up to 1000 for other cases). Use q < 0.05 for robust analyses; q < 0.25 is called very lenient. · Step 2A: Pathway enrichment analysis of a gene…; Step 2B: Pathway enrichment analysis of a ranked…; General recommendations for pathway enrichment…; Part 1B - Pathway enrichment analysis of a…

  5. ClaimedEarlier applications are cited as motivation. Pathway analysis pointed to PRC2-driven histone and DNA methylation as a therapeutic target in ependymoma. It also found deletion-affected pathways in rare autism CNVs where single-gene tests gave few hits.

    Summarized from earlier work, with no data in this text: 5-azacytidine, given on a compassionate basis to a terminally ill patient, reportedly stopped rapid metastatic growth. Illustration: a list with 40% cell cycle genes is enriched when cell cycle genes make up only 8% of human protein-coding genes. · INTRODUCTION

Methods, data and reproducibility

Three-stage workflow: derive a gene list or genome-wide ranked list from omics data (e.g., edgeR or limma differential expression), test enrichment with g:Profiler (Fisher's exact or ordered test) or GSEAPreranked (permutation-based modified Kolmogorov-Smirnov), then cluster overlapping pathways into themes in Cytoscape Enrichment Map.

  • TCGA exome sequencing: genes with frequent somatic SNVs; the 127 cancer drivers called by MuSiC are the g:Profiler input (Supplementary Table 1) · human (TCGA tumors) · n = 3,200 tumors of 12 types; 127 driver genes
  • TCGA ovarian cancer RNA-seq ranked gene list contrasting mesenchymal vs immunoreactive subtypes (GSEA input, Supplementary Table 2) · human (TCGA tumors)
  • TCGA serous ovarian cancer Affymetrix microarray data, RMA normalized, from the GDC portal (Supplementary Protocol 1A) · human (TCGA tumors)
  • TCGA ovarian cancer RNA-seq RSEM read counts with subtype class definitions, analysed with edgeR (Supplementary Protocol 1B) · human (TCGA tumors)
  • Bader lab pathway GMT (GO BP without IEA, Reactome, Panther, NetPath, NCI, MSigDB, HumanCyc) downloaded July 1, 2017 (Supplementary Table 3) · human
  • g:Profiler Homo sapiens GMT of GO biological process and Reactome gene sets (Supplementary Table 5) · human
Methods & tools
g:Profiler (Fisher's exact test, ordered query; g:Convert), GSEA desktop (GSEAPreranked, gene set permutation), Cytoscape (3.5.1 or higher), Enrichment Map Cytoscape app (3.0 or higher), AutoAnnotate (1.2.0 or higher), clusterMaker2 (0.9.5 or higher), WordCloud (3.1.0 or higher), edgeR, limma, MuSiC (cancer driver gene detection), GO, Reactome, Panther, NetPath, NCI PID, MSigDB, HumanCyc gene sets, R/Bioconductor: Biobase, GSA, gProfileR, pheatmap
Software & environment
Java SE 8 or higher; GSEA desktop 3.0; Cytoscape 3.5.1 or higher with Enrichment Map 3.0+, clusterMaker2 0.9.5+, WordCloud 3.1.0+, AutoAnnotate 1.2.0+; R/RStudio with Bioconductor Biobase, limma, edgeR, GSA, locfit; gene sets dated July 1, 2017.
Assay conditions & compute
PC with at least 8GB RAM (1GB suffices for GSEA); about 3 min for g:Profiler, 20 min for GSEA (Windows7, 1GB RAM, Java 8), 5 min to build and 4 h to annotate the map; 1000 permutations took 1012 s on a 16G RAM i7 3.4 GHz machine.
Reproducibility gaps
Example inputs ship as supplementary tables rather than through a stated public repository; GSEA download requires free registration; phenotype-permutation GSEA for RNA-seq needs custom code and is not provided.
Limitations (6)
  • Pathways regulated outside gene expression, such as by post-translational control, go undetected. Treating pathways as gene sets also discards interactions, biochemical reactions, modifications, complexes and activation or inhibition relationships. · stated by the authors
  • Results are biased toward well-studied pathways, and genes without annotations are ignored. Highly ranked multifunctional genes can drive enrichment of many pathways that are irrelevant to the experiment. · stated by the authors
  • Most tests assume genes and pathways are independent, despite co-expression and shared members, so BH-FDR (Benjamini-Hochberg false discovery rate) estimates may be too liberal or too conservative. Bonferroni correction is considered overly strict. · stated by the authors
  • Only GSEA gene set permutation is covered. Phenotype permutation, which models gene-gene correlation and suits larger replicate counts, needs custom programming for RNA-seq and is left out. · stated by the authors
  • Standard tests are inflated by genomically clustered families (olfactory receptors, histones, MHC, homeobox genes). ChIP-seq regions and large CNV or GWAS intervals need specialized tools such as GREAT, PLINK or INRICH. · stated by the authors
  • No benchmark compares g:Profiler, GSEA or alternatives (e.g., Camera) on sensitivity or false discovery. The recommended size limits and q-value cutoffs rest on expert judgment, not on data shown in the text. · noted in this summary

Read this if you need a hands-on, parameter-level guide to running g:Profiler or GSEA on omics gene lists and turning the enriched pathways into annotated Enrichment Map figures in Cytoscape.

Citations
1,972
FWCI
69.87
Field percentile
top 1%

Abstract

Pathway enrichment analysis helps gain mechanistic insight into large gene lists typically resulting from genome scale (-omics) experiments. It identifies biological pathways that are enriched in the gene list more than expected by chance. We explain pathway enrichment analysis and present a practical step-by-step guide to help interpret gene lists resulting from RNA-seq and genome sequencing experiments. The protocol comprises three major steps: define a gene list from genome scale data, determine statistically enriched pathways, and visualize and interpret the results. We focus on differentially expressed genes and mutated cancer genes, however the described principles can be applied to diverse -omics data. The protocol is designed for biologists with no prior bioinformatics training and uses freely available software including g:Profiler, GSEA, Cytoscape and Enrichment Map.

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISRegmi '23García-Mili '26Wang '26Hodge '23Canzler '20Candia '24
63% semantic matchBioinformatics
Binod Regmi, Stephen R. Brooks, ANDREW J. UHLMAN+1 more·bioRxiv·Aug 28, 2023