How to Choose the Right Gene Set for Enrichment Analysis: GO, KEGG, Reactome, or Custom Gene Sets?

By Dr. Zubair Khalid, DVM, MS, PhD · 2026-08-20

Key Takeaways

• The choice of gene set collection (GO, KEGG, Reactome, or custom) is dictated by the specific biological question, with GO offering broad functional annotation, KEGG mapping to metabolic/signaling pathways, Reactome detailing molecular reaction sequences, and custom sets testing hypothesis-driven queries.
• Annotation quality and species coverage are critical; GO annotations vary in evidence type (experimental vs. computational), KEGG pathway completeness differs by organism, Reactome is most robust for human biology, and custom sets require rigorous self-curation.
• Statistical methodology (Over-Representation Analysis vs. Functional Class Scoring like GSEA) interacts with gene set choice; ORA relies on thresholded gene lists, while GSEA uses full gene rankings, making the ranking metric selection paramount for GSEA.
• Practical implementation involves defining research questions, assessing data type/quality, verifying gene set coverage for the organism, selecting an appropriate statistical method, and comparing results across multiple collections for robustness.
• Reproducibility hinges on meticulous documentation, including database versions (e.g., KEGG release), gene identifier mapping, statistical parameters, and reporting gene set coverage rates to enable validation and interpretation.
• Common failure patterns include using incorrect background lists in ORA, ignoring gene set size effects, selecting inappropriate ranking metrics for GSEA, overinterpreting statistical significance without biological validation, and neglecting multiple testing correction.

Gene set enrichment analysis is a standard step after differential expression testing in RNA sequencing workflows. The choice of gene set collection determines which biological questions your analysis can answer. Gene Ontology (GO) terms describe molecular functions, cellular components, and biological processes. KEGG pathways map genes to curated metabolic and signaling routes. Reactome organizes reactions into hierarchical pathways. Custom gene sets let you test hypotheses that curated databases do not cover. This article provides a decision framework for selecting among these collections based on your research goals, organism, data type, and statistical method. The framework applies to students, researchers, and laboratory professionals who perform enrichment analysis as part of RNA-seq, microarray, proteomics, or metabolomics studies.

The Core Decision: What Biological Question Are You Asking?

The gene set collection you choose should match the biological question you want to answer. Each major collection has a distinct organizing principle, and these principles lead to different interpretations of your results.

GO is structured as a directed acyclic graph with three independent ontologies. The molecular function ontology describes activities at the molecular level, such as catalytic activity or binding. The cellular component ontology describes locations within the cell, such as the mitochondrion or the plasma membrane. The biological process ontology describes series of events accomplished by one or more ordered assemblies of molecular functions. When you test GO terms, you ask whether your gene list is enriched for genes that share a functional annotation, a cellular location, or a process-level role.

KEGG pathways are manually curated maps of molecular interaction and reaction networks. These pathways cover metabolism, genetic information processing, environmental information processing, cellular processes, organismal systems, and human diseases. When you test KEGG pathways, you ask whether your genes converge on a known interaction network, often with directional information about which reactions feed into which products.

Reactome is a free, open-source, curated pathway database that organizes human biological processes into hierarchical pathways. The database covers reactions, which are the fundamental units, and groups them into pathways and higher-level categories. Reactome includes detailed molecular information such as protein modifications, complex assembly, and translocation events. When you test Reactome pathways, you ask whether your genes participate in a defined sequence of molecular reactions.

Custom gene sets are collections you define yourself. These can come from published gene signatures, results of previous experiments, genomic intervals, or literature curation. When you test custom gene sets, you ask whether your data supports a hypothesis that is specific to your study and not captured by general-purpose databases.

The practical consequence of these differences is that the same gene list can produce different enrichment results depending on the collection used. A gene list from a cancer study might show significant enrichment for GO biological processes related to cell cycle regulation, KEGG pathways related to p53 signaling, and Reactome pathways related to cell cycle checkpoints. Each result is correct within its own annotation framework, but each answers a slightly different question about the biology.

At a Glance: Gene Set Collection Comparison

Core Principles of Gene Set Selection

Annotation Quality and Curation Status

The reliability of your enrichment results depends on the quality of the underlying annotations. GO annotations are assigned through a combination of experimental evidence, computational prediction, and curator inference. Each annotation includes an evidence code that indicates how the annotation was made. Experimental evidence codes are generally more reliable than computational ones. When you interpret GO enrichment results, you should consider whether the significant terms are supported by experimental annotations or primarily by computational predictions.

KEGG pathways are manually curated from published literature. The curation process involves reading primary research articles and extracting molecular interaction information. This manual process means KEGG pathways are generally high quality, but they can lag behind the current literature. Pathway updates occur regularly, and the version of the KEGG database you use affects your results. The clusterProfiler package can directly access the latest KEGG knowledge database, which helps ensure you are testing against current pathway definitions.

Reactome uses a peer-reviewed curation model. Curators extract reaction details from literature and experts review the annotations. The database includes detailed molecular information such as post-translational modifications and subcellular localization. This depth of annotation makes Reactome particularly useful for mechanistic studies, but the curation burden means coverage varies across biological areas.

Custom gene sets have variable quality depending on their source. Published gene signatures from high-quality studies can be reliable, but you need to verify the original evidence. Gene sets derived from computational predictions or automated text mining require additional validation. The key principle is that you should document the source and evidence for every custom gene set you use.

Species Coverage and Orthology Mapping

The species you study determines which gene set collections are usable. GO annotations exist for thousands of species, but the depth of annotation varies enormously. Model organisms such as human, mouse, zebrafish, and yeast have extensive experimental annotations. Non-model organisms often have annotations that are inferred from orthology to model organisms, which introduces uncertainty.

KEGG pathways are organized by species-specific pathway maps. The KEGG database includes pathway maps for many organisms, but the completeness of these maps varies. Some organisms have well-curated metabolic pathways, while others have sparse coverage. When you work with a species that has limited KEGG coverage, you may need to map your genes to a closely related reference species and interpret the results with caution.

Reactome has the most comprehensive coverage for human biology. The database includes orthology-based projections to other species, but these projections are less detailed than the human annotations. For non-human species, you should check the number of genes from your dataset that map to Reactome pathways. Low mapping rates indicate that Reactome may not be the best choice for your species.

Custom gene sets solve the species coverage problem because you define the gene members. If you study a non-model organism, you can create gene sets based on your own knowledge or on orthology mapping from model organisms. The tradeoff is that you take responsibility for the accuracy of the gene-to-set assignments.

Statistical Method Compatibility

The statistical method you use for enrichment analysis interacts with your choice of gene set collection. Over-representation analysis (ORA) tests whether the genes in your significant list are enriched for a particular annotation compared to what you would expect by chance. ORA requires a defined list of significant genes, which means you need to choose a significance threshold for your differential expression results. The choice of threshold affects which gene sets appear enriched.

Functional class scoring (FCS) methods, such as gene set enrichment analysis (GSEA), use the full ranking of genes instead of a thresholded list. GSEA ranks all genes by a metric such as fold change or test statistic, then tests whether genes in a set are randomly distributed throughout the ranking or concentrated at the top or bottom. The choice of ranking metric has a critical impact on the results of pathway enrichment analysis. Research comparing 16 ranking metrics across 28 benchmark datasets found that the absolute value of the Moderated Welch Test statistic and the absolute value of the Signal-To-Noise ratio gave stable results across sample sizes. Applying a default ranking metric without considering your data structure may lead to poor results.

The gene set collection you choose affects both ORA and GSEA. For ORA, the size of the gene sets matters. Very small gene sets may not have enough genes in your significant list to produce meaningful enrichment. Very large gene sets may produce significant results that are biologically uninformative because they encompass too many processes. For GSEA, the distribution of gene set sizes affects the statistical power. Methods that account for varying set sizes when testing multiple gene sets tend to perform better than methods that ignore size differences.

Practical Workflow for Gene Set Selection

Step 1: Define Your Research Question and Hypothesis

Write down the specific biological question you want to answer. Examples include identifying metabolic pathways that differ between treatment groups, determining which biological processes are disrupted in a disease state, or testing whether a published gene signature is active in your dataset. The specificity of your question guides the choice of gene set collection.

For broad exploratory questions, GO biological process terms provide a general survey of affected biology. For questions about specific metabolic or signaling mechanisms, KEGG pathways are more appropriate. For questions about detailed molecular mechanisms, Reactome offers the most granular information. For hypothesis-driven questions based on prior knowledge, custom gene sets are the only option that directly tests your specific hypothesis.

Step 2: Assess Your Data Type and Quality

The type of data you have affects which gene set collections are usable. RNA-seq data provides gene-level expression measurements that map directly to gene sets. Microarray data also maps to gene sets, but the probe-to-gene mapping requires careful handling. Proteomics and metabolomics data can be analyzed with gene set methods, but the coverage of the detected molecules affects the results. Spatial transcriptomics data requires computational solutions that account for the spatial organization of gene expression, and the choice of gene set collection should consider the platform limitations.

Data quality affects enrichment results. Low-quality data with high noise levels can produce spurious enrichment signals. Before running enrichment analysis, verify that your differential expression results are robust. Check the distribution of p-values, the magnitude of effect sizes, and the consistency of results across biological replicates. The choice of ranking metric for GSEA should be based on your data characteristics, including sample size and the distribution of expression values.

Step 3: Check Gene Set Coverage for Your Organism

Before committing to a gene set collection, calculate the proportion of your measured genes that map to the collection. Most enrichment tools report this mapping rate. A low mapping rate means that many of your genes are not represented in the collection, which reduces statistical power and can bias results.

For GO, check the number of annotated genes for your species. For KEGG, check whether your species has dedicated pathway maps. For Reactome, check the orthology-based coverage. If mapping rates are low for curated databases, consider using custom gene sets based on orthology mapping from a well-annotated reference species.

Step 4: Select the Statistical Method

Decide between ORA and FCS based on your data and question. ORA is appropriate when you have a clear list of significant genes and want to know which annotations are over-represented. FCS methods such as GSEA are appropriate when you want to use the full ranking of genes and avoid arbitrary significance thresholds.

The choice of ranking metric for GSEA requires attention. Research has shown that the choice of ranking metric has a critical impact on the results of pathway enrichment analysis. The absolute value of the Moderated Welch Test statistic and the absolute value of the Signal-To-Noise ratio provided stable results across sample sizes in benchmark testing. Other metrics performed better with larger sample sizes. You should test multiple ranking metrics and compare the consistency of your results.

Step 5: Run the Analysis and Compare Collections

Run enrichment analysis with multiple gene set collections to understand how your results depend on the annotation framework. This comparison is informative even when one collection is your primary choice. Differences in results across collections can reveal whether your findings are robust or specific to a particular annotation system.

The clusterProfiler package supports multiple gene set collections and provides visualization options that aid interpretation. The package supports GO and KEGG through over-representation and gene set enrichment analyses. It also supports custom gene sets. The computational steps for typical analyses can be completed in about two minutes, which makes it practical to compare multiple collections.

Step 6: Interpret Results in Biological Context

Interpret enrichment results in the context of your experimental design and biological knowledge. Significant enrichment does not prove biological causation. A gene set can be enriched because of coordinated regulation, because of shared regulatory elements, or because of technical artifacts. Consider whether the enriched terms make biological sense given your experimental system.

The interpretation of GSEA results requires attention to the distinction between statistical significance and biological distinction. Research comparing pairwise GSEA with single-sample approaches found that pairwise GSEA can overgeneralize biological enrichment. When the most statistically significant signatures were assessed using single-sample approaches, there was a complete absence of biological distinction between groups. This finding means you should validate your enrichment results with single-sample methods or other independent approaches before drawing biological conclusions.

Gene Ontology: Strengths and Use Cases

Structure and Annotation Depth

GO provides a controlled vocabulary for describing gene products across three domains. The molecular function ontology describes what a gene product does at the biochemical level. The cellular component ontology describes where a gene product is active. The biological process ontology describes the larger processes to which a gene product contributes.

The structure of GO as a directed acyclic graph means that terms are related through parent-child relationships. A gene annotated to a specific child term is also implicitly annotated to all parent terms. This structure creates redundancy in enrichment results, where multiple related terms appear significant. Tools such as clusterProfiler provide functions to reduce this redundancy and identify representative terms.

When to Use GO

Use GO when you need a broad functional survey of your gene list. GO is particularly useful for hypothesis generation because it covers a wide range of biological functions. If you have a list of differentially expressed genes and want to know what kinds of processes they participate in, GO biological process terms provide a starting point.

GO is also useful for non-model organisms because annotations can be inferred from orthology. The quality of these inferred annotations varies, but they provide a starting point for interpretation. The NCBI provides access to sequence resources and analysis services that can help with orthology mapping and functional annotation.

Limitations of GO

The main limitation of GO is redundancy. The directed acyclic graph structure means that significant terms often include many parent and child terms that describe the same biology. This redundancy can make results difficult to interpret and can obscure the most relevant biological signal.

GO annotations also vary in quality. Experimental annotations are more reliable than computational predictions. When you interpret GO enrichment results, check the evidence codes for the annotations that drive the significant terms. If the significant terms are supported primarily by computational annotations, your confidence in the biological interpretation should be lower.

KEGG Pathways: Strengths and Use Cases

Pathway Structure and Curation

KEGG pathways are manually curated maps of molecular interaction networks. Each pathway includes genes, proteins, and small molecules connected by reactions and interactions. The pathways are organized into categories that reflect biological function, including metabolism, genetic information processing, and environmental information processing.

The manual curation of KEGG pathways means they capture established biological knowledge. The pathways are particularly strong for metabolism, where the reaction networks are well characterized. KEGG also includes disease pathways that connect molecular mechanisms to human diseases.

When to Use KEGG

Use KEGG when you want to interpret your results in the context of known metabolic or signaling pathways. KEGG is particularly useful for studies of metabolism, where the pathway maps provide a framework for understanding how changes in gene expression affect biochemical networks.

KEGG is also useful for disease mechanism studies. The disease pathways in KEGG connect genes to disease processes, which can help you interpret the clinical relevance of your findings. Studies of chronic obstructive pulmonary disease have used KEGG analysis to explore potential mechanisms and identify biomarkers.

Limitations of KEGG

The main limitation of KEGG is species coverage. While KEGG includes pathway maps for many organisms, the completeness varies. Some organisms have extensive pathway coverage, while others have sparse maps. For species with limited coverage, you may need to map genes to a reference species and interpret results with caution.

KEGG pathway updates also affect reproducibility. The database is updated regularly, and the version you use affects your results. Document the KEGG version in your analysis records so that others can reproduce your results. The clusterProfiler package can directly access the latest KEGG knowledge database, which helps ensure you are testing against current pathway definitions.

Reactome: Strengths and Use Cases

Reaction-Based Pathway Organization

Reactome organizes biological knowledge into reactions, which are the fundamental units of the database. Reactions describe molecular transformations, including binding, catalysis, and translocation. Reactions are grouped into pathways, and pathways are organized into hierarchical categories.

The reaction-based organization provides detailed mechanistic information. Reactome includes information about protein modifications, complex assembly, and subcellular localization. This detail makes Reactome particularly useful for understanding the molecular mechanisms underlying your observations.

When to Use Reactome

Use Reactome when you need detailed mechanistic information about molecular reactions. Reactome is particularly useful for studies of signaling pathways, where the sequence of reactions matters for interpretation. The hierarchical organization of Reactome also allows you to examine results at different levels of detail, from broad pathway categories to specific reactions.

Reactome is most comprehensive for human biology. If you study human cells or tissues, Reactome provides the most detailed annotations. For other species, Reactome uses orthology-based projections that are less detailed than the human annotations.

Limitations of Reactome

The main limitation of Reactome is species coverage. The database is most comprehensive for human biology, and coverage for other species depends on orthology mapping. For non-human species, check the mapping rate before using Reactome as your primary gene set collection.

Reactome pathway sizes can also be large, which can dilute statistical signal. Large pathways that encompass many genes may not show significant enrichment even when a relevant subset of genes is affected. Consider using the hierarchical structure of Reactome to test more specific sub-pathways.

Custom Gene Sets: Strengths and Use Cases

Defining Custom Collections

Custom gene sets are collections you define based on your own knowledge, published literature, or previous experiments. These sets can be derived from gene signatures in published studies, from genomic intervals identified in your own experiments, or from manual curation of the literature.

The clusterProfiler package supports custom gene sets, as do other enrichment tools. You provide the gene-to-set mapping in a format the tool accepts, and the tool tests your sets for enrichment. The flexibility of custom gene sets makes them valuable for hypothesis-driven research.

When to Use Custom Gene Sets

Use custom gene sets when you have a specific hypothesis that curated databases do not cover. Examples include testing whether a published gene signature is active in your dataset, testing whether genes from a specific genomic region are coordinately regulated, or testing whether a set of genes you identified in a previous experiment replicates in a new dataset.

Custom gene sets are also useful for non-model organisms. If your species has limited coverage in GO, KEGG, or Reactome, you can create custom gene sets based on orthology mapping from a well-annotated reference species. This approach lets you test biological hypotheses even when curated databases lack coverage for your organism.

Limitations of Custom Gene Sets

The main limitation of custom gene sets is the risk of circular analysis. If you define gene sets based on the same data you are testing, your enrichment results are not independent. This circularity can produce inflated significance. To avoid this problem, define your custom gene sets based on external information, such as published literature or independent experiments.

Custom gene sets also require careful documentation. You need to record the source of each gene set, the evidence supporting the gene-to-set assignments, and the version of the set you used. This documentation is essential for reproducibility and for others to evaluate the validity of your results.

Practical Implementation Steps

Setting Up Your Analysis Environment

The Bioconductor project provides packages for enrichment analysis, including clusterProfiler. The project provides official documentation for package installation and workflow execution. You can install Bioconductor packages using the standard installation procedures described in the project documentation.

The Galaxy Training Network provides accessible workflow training for bioinformatics analysis. The training materials cover enrichment analysis and related topics. These resources are useful for researchers who prefer graphical interfaces over command-line tools.

The Carpentries provides foundational computing lessons that cover shell, Git, and programming skills. These skills are useful for managing your analysis workflow and ensuring reproducibility. The nf-core documentation describes community standards for reproducible workflows, which can help you structure your analysis.

Preparing Your Gene List

The format of your gene list depends on the enrichment tool you use. Most tools accept gene identifiers such as Entrez Gene IDs, Ensembl IDs, or gene symbols. The NCBI provides access to sequence resources and search systems that can help you convert between identifier types.

For ORA, you need a list of significant genes and a background list. The background list should include all genes that were tested for differential expression, beyond the significant ones. Using the wrong background can produce biased enrichment results.

For GSEA, you need a ranked list of all genes. The ranking metric should be chosen based on your data characteristics. Research has shown that the choice of ranking metric has a critical impact on the results of pathway enrichment analysis. Test multiple ranking metrics and compare the consistency of your results.

Running the Enrichment Analysis

The clusterProfiler package provides functions for both ORA and GSEA. The package supports GO, KEGG, Reactome, and custom gene sets. The package also provides visualization functions that help you interpret your results.

For GO analysis, you specify the ontology you want to test. You can test molecular function, cellular component, or biological process separately. The package provides functions to reduce redundancy in GO results and identify representative terms.

For KEGG analysis, you specify the organism code for your species. The package can directly access the latest KEGG knowledge database. For Reactome analysis, you use the ReactomePA package or the clusterProfiler interface.

For custom gene sets, you provide the gene-to-set mapping. The format depends on the tool you use. The clusterProfiler package accepts gene set collections in standard formats.

Visualizing and Interpreting Results

Visualization is essential for interpreting enrichment results. The clusterProfiler package provides various graphical outputs, including dot plots, term-gene network plots, enrichment map plots, ridge plots, and GSEA plots. These visualizations help you understand which gene sets are enriched and which genes drive the enrichment.

Dot plots show the enrichment score and gene count for each significant term. Term-gene network plots show the connections between terms and genes. Enrichment map plots show the relationships between enriched terms. Ridge plots show the distribution of gene-level statistics for each set. GSEA plots show the running enrichment score and the gene ranking.

The choice of visualization depends on your question. Dot plots are useful for comparing enrichment across terms. Network plots are useful for understanding the relationships between terms. GSEA plots are useful for understanding which genes drive the enrichment signal.

Records and Measurements

Documenting Your Analysis

Reproducibility requires careful documentation of your enrichment analysis. Record the version of every tool and database you use. Record the gene identifiers and the mapping to gene sets. Record the statistical parameters, including the ranking metric, the number of permutations, and the significance thresholds.

The Bioconductor project provides documentation for reproducible genomic analysis. The nf-core documentation describes community standards for reproducible workflows. These resources can help you structure your documentation.

Measuring Gene Set Coverage

Gene set coverage is a key quality metric. Calculate the proportion of your measured genes that map to each gene set collection. Low coverage indicates that the collection may not be appropriate for your data. Report the coverage in your methods so that others can evaluate the reliability of your results.

For GO, coverage depends on the annotation depth for your species. For KEGG, coverage depends on the availability of species-specific pathway maps. For Reactome, coverage depends on orthology-based projections. For custom gene sets, coverage depends on the quality of your gene-to-set assignments.

Tracking Enrichment Results

Keep a record of your enrichment results, including the significant terms, the enrichment scores, and the adjusted p-values. This record allows you to compare results across analyses and to evaluate the consistency of your findings.

If you run multiple gene set collections, record the results for each collection. Comparing results across collections can reveal whether your findings are robust or specific to a particular annotation framework.

Common Failure Patterns

Using the Wrong Background List

A common error in ORA is using an incorrect background list. The background should include all genes that were tested for differential expression, beyond the significant ones. Using a restricted background can produce inflated enrichment scores because the expected frequency of annotations is underestimated.

Ignoring Gene Set Size Effects

Gene set size affects enrichment results. Very small gene sets may not have enough genes to produce meaningful enrichment. Very large gene sets may produce significant results that are biologically uninformative. Methods that account for varying set sizes when testing multiple gene sets tend to perform better than methods that ignore size differences.

Choosing an Inappropriate Ranking Metric

The choice of ranking metric for GSEA has a critical impact on the results of pathway enrichment analysis. Applying a default ranking metric may lead to poor results. Research comparing 16 ranking metrics across 28 benchmark datasets found that the absolute value of the Moderated Welch Test statistic and the absolute value of the Signal-To-Noise ratio gave stable results across sample sizes. Test multiple ranking metrics and compare the consistency of your results.

Overinterpreting Statistical Significance

Statistical significance in enrichment analysis does not prove biological relevance. A gene set can be enriched because of coordinated regulation, because of shared regulatory elements, or because of technical artifacts. Research comparing pairwise GSEA with single-sample approaches found that pairwise GSEA can overgeneralize biological enrichment. Validate your enrichment results with independent methods before drawing biological conclusions.

Failing to Account for Multiple Testing

Enrichment analysis tests many gene sets simultaneously, which requires multiple testing correction. The choice of correction method affects your results. Common methods include Benjamini-Hochberg false discovery rate control and family-wise error rate control. The simultaneous enrichment analysis approach uses closed testing with Simes tests to control the family-wise error rate. Choose a correction method appropriate for your analysis and report it in your methods.

Limitations and Interpretation Boundaries

Annotation Incompleteness

All gene set collections are incomplete. GO annotations are missing for many genes, particularly in non-model organisms. KEGG pathways do not cover all biological processes. Reactome has limited coverage outside human biology. Custom gene sets are limited by your own knowledge and the quality of your sources.

Annotation incompleteness means that absence of enrichment does not prove absence of biological effect. A gene set may not appear enriched because the annotations are incomplete, not because the biology is unaffected. Interpret negative results with caution.

Database Version Dependence

Enrichment results depend on the version of the gene set database you use. GO, KEGG, and Reactome are updated regularly. The version you use affects your results. Document the database versions in your methods so that others can reproduce your results.

Platform Differences

The platform you use for gene expression measurement affects enrichment results. Research comparing next-generation sequencing and microarrays found higher reproducibility on microarray data for gene ranking, but similar enrichment of top gene sets at the pathway level. The choice of platform affects the genes you detect and the accuracy of their expression measurements.

Multi-Omics Integration

Enrichment analysis is increasingly applied to multi-omics data. Integrating transcriptomics, proteomics, and metabolomics data requires methods that account for the different molecular layers. The clusterProfiler package supports multi-omics enrichment analysis. Protocols for multi-omics integration describe steps for data preparation, model construction, and biological interpretation.

Safety and Regulatory Context

Data Management and Privacy

Gene expression data from human subjects may be subject to privacy regulations. Ensure that your data handling procedures comply with applicable regulations. The NCBI provides access to sequence resources and analysis services, but you are responsible for ensuring that your data submission and analysis comply with ethical and legal requirements.

Reproducibility Standards

Funding agencies and journals increasingly require reproducible analysis. The Bioconductor project provides documentation for reproducible genomic analysis. The nf-core documentation describes community standards for reproducible workflows. The Galaxy Training Network provides accessible workflow training. Following these standards helps ensure that your enrichment analysis can be reproduced by others.

Professional Escalation Criteria

If your enrichment results are inconsistent across gene set collections, or if the results conflict with established biological knowledge, consider seeking advice from a bioinformatics specialist. Inconsistency across collections may indicate problems with your gene list, your ranking metric, or your statistical method. A specialist can help you diagnose the problem and identify appropriate solutions.

If your custom gene sets produce results that are difficult to interpret, consider whether the gene-to-set assignments are correct. Review the evidence supporting each assignment and consider whether the sets are appropriately sized. A specialist can help you evaluate the quality of your custom gene sets.

Frequently Asked Questions

What is the difference between over-representation analysis and gene set enrichment analysis?

Over-representation analysis tests whether a defined list of significant genes contains more genes from a particular gene set than expected by chance. This method requires a significance threshold to define the gene list. Gene set enrichment analysis uses the full ranking of genes and tests whether genes in a set are concentrated at the top or bottom of the ranking. GSEA does not require a significance threshold and can detect coordinated changes in gene expression that are too small to reach individual significance.

How do I choose between GO, KEGG, and Reactome for my analysis?

Choose based on your biological question. GO provides a broad functional survey across molecular function, cellular component, and biological process. KEGG provides curated pathway maps that are particularly strong for metabolism and signaling. Reactome provides detailed reaction-based pathways that are most comprehensive for human biology. For hypothesis-driven questions, custom gene sets are the only option that directly tests your specific hypothesis.

Can I use multiple gene set collections in one analysis?

Yes. Running enrichment analysis with multiple collections can reveal whether your findings are robust or specific to a particular annotation framework. The clusterProfiler package supports multiple collections and provides visualization options that aid comparison. Differences in results across collections can provide biological insight.

What should I do if my organism has poor coverage in curated databases?

Check the mapping rate for each collection. If coverage is low, consider using custom gene sets based on orthology mapping from a well-annotated reference species. You can also use GO annotations that are inferred from orthology, but interpret results with caution because inferred annotations are less reliable than experimental ones.

How do I choose the ranking metric for GSEA?

The choice of ranking metric has a critical impact on the results of pathway enrichment analysis. Research comparing 16 ranking metrics found that the absolute value of the Moderated Welch Test statistic and the absolute value of the Signal-To-Noise ratio gave stable results across sample sizes. Test multiple ranking metrics and compare the consistency of your results.

What is the difference between self-contained and competitive gene set tests?

Self-contained tests ask whether the genes in a set are differentially expressed without comparing to genes outside the set. Competitive tests ask whether the genes in a set are more differentially expressed than genes outside the set. Both approaches have advantages and disadvantages. Simultaneous enrichment analysis provides a unified approach that includes both self-contained and competitive null hypotheses as special cases.

How do I avoid circular analysis with custom gene sets?

Define your custom gene sets based on external information, such as published literature or independent experiments. Do not define gene sets based on the same data you are testing. Document the source of each gene set and the evidence supporting the gene-to-set assignments.

How do I validate my enrichment results?

Validate your results with independent methods. Use single-sample approaches such as single-sample GSEA or gene set variation analysis to check whether the enrichment signal is consistent across individual samples. Compare results across gene set collections. Consider whether the enriched terms make biological sense given your experimental system.

Related Bioinformatics Guides

• Gene Set Enrichment Analysis Tools: Choosing the Right One
• How to Interpret Gene Set Enrichment Analysis Results
• Gene Set Enrichment Analysis in R: A Practical Tutorial for Interpreting Omics Data
• Gene Ontology (GO) and Enrichment Analysis
• Pathway Enrichment Analysis for Proteomics: Tools and Interpretation

Related Clinical & Scientific Guides

• A Practical Guide to Detecting Antimicrobial Resistance Genes in Shotgun Metagenomic Data
• Computational Immunology: Modeling the Immune System
• How to Set Hard Filters for Germline Variant Calling: A Practical Guide to GATK Best Practices

References and Further Reading

• NCBI Data Resources. National Center for Biotechnology Information.
• EMBL-EBI Training. European Bioinformatics Institute.
• Bioconductor. Bioconductor Project.
• Galaxy Training Network. Galaxy Project.
• nf-core Documentation. nf-core.
• The Carpentries Lessons. The Carpentries.
• Using clusterProfiler to characterize multiomics data.. Nature protocols, 2024.
• Ranking metrics in gene set enrichment analysis: do they matter?. BMC bioinformatics, 2017.
• Simultaneous Enrichment Analysis of all Possible Gene-sets: Unifying Self-Contained and Competitive Methods.. Briefings in bioinformatics, 2020.
• Gene set enrichment analysis made simple.. Statistical methods in medical research, 2009.
• Meta-analysis approaches to combine multiple gene set enrichment studies.. Statistics in medicine, 2018.
• Gene set enrichment meta-learning analysis: next- generation sequencing versus microarrays.. BMC bioinformatics, 2010.
• Dual gene set enrichment analysis (dualGSEA), an R function that enables more robust biological discovery and pre-clinical model alignment from transcriptomics data.. Scientific reports, 2024.
• Computational solutions for spatial transcriptomics.. Computational and structural biotechnology journal, 2022.
• Protocol for integrating and interpreting multi-omics data combining unsupervised and supervised data integrating approaches.. 2026.
• Protocol for analyzing potential targets of environmental pollutants in human diseases using network toxicology and molecular docking.. 2026.
• Biological Functional Class Enrichment Analysis with R, an Annotated Tutorial for Bench Scientists.. 2026.
• CGPS: A machine learning-based approach integrating multiple gene set analysis tools for better prioritization of biologically relevant pathways.. Journal of genetics and genomics = Yi chuan xue bao, 2018.
• Identification of Inflammation-Related Biomarker Lp-PLA2 for Patients With COPD by Comprehensive Analysis. Frontiers in Immunology, 2021.
• Irinotecan's molecular mechanisms against cancer: a primary system biology and chemoinformatics approach for novel formulation development. Current Issues in Pharmacy and Medical Sciences, 2025.
• Pathway Enrichment Analysis of Microarray Data. Methods in Molecular Biology, 2022.
• Mining functionally relevant gene sets for analyzing physiologically novel clinical expression data. Pacific Symposium on Biocomputing 2011 Psb 2011, 2011.

This article is educational and does not replace validated analysis plans, institutional policy, clinical interpretation, or specialist review.