Why Background Gene Set Selection Matters in Pathway and Enrichment Analysis

In bioinformatics, careful attention to detail is critical when analyzing gene expression and protein expression data. One of the most frequent questions we hear is: What background should be used when performing pathway or other gene set enrichment analysis? Let’s unpack why this choice matters and how using the right background ensures accurate, reproducible results.

Before we explore why background selection matters, it’s important to clarify what the background (or reference) gene set is. In pathway and gene set enrichment analysis, the background defines the full set of genes or proteins considered eligible for statistical comparison. It represents all the genes that could have been detected as significant in your experiment, and it’s used to calculate whether the observed overlap between your differentially expressed genes (DEGs) and a given pathway is greater than what would be expected by chance.

Here’s a helpful way to think about it:

Imagine a raffle where your odds of winning depend on how many tickets are in the bowl. If there are 100 tickets and you hold 10, your chances are fairly high. But if someone increases the total number of tickets to 1,000 without changing how many you hold, your chances of winning go down. Enrichment analysis works the same way. The size and contents of your background set affect how significant the overlap with a pathway appears. Using an inflated or unrelated background can make your results seem more or less significant than they truly are.

Your background should always match the full list of genes or proteins that were actually measured in your experiment. This ensures that statistical results are accurate and reflect the true context of your data.

Choosing an appropriate background is critical because it directly influences the p-values and perceived significance of enriched pathways. These values reflect the probability that the observed overlap between your gene list and a pathway occurred by chance, given the background you’ve selected.

Choosing an inappropriate background, like all genes from a public database instead of only those measured in your experiment, can inflate the number of significant results and distort your results. It increases the risk of false positives and makes it harder to trust the conclusions from your analysis.

When you use the correct background, you’re grounding your statistical tests in the actual experimental context, ensuring that your results are both accurate and reproducible.

To illustrate the impact of background selection, we analyzed 1,172 differentially expressed genes (FDR adjusted p < 0.05 and log₂ fold change > |0.6|) from a published dataset (GEO accession: GSE47363). We performed enrichment analysis twice using two different backgrounds:

1. All the genes measured in the experiment, approximately 20,000 genes.
2. An arbitrary set of approximately 30,000 genes from the NCBI human gene database, as used by some popular pathway analysis tools.

The results were striking. When using the measured gene pool as background, we identified 64 significant pathways with false discovery rate (FDR) p-values below 0.05. Using the larger, arbitrary NCBI gene pool as background, the number of significant pathways more than doubled to over 150, and the p-values appeared overly significant, indicating false positives.

While the top pathway remained consistent between the two analyses, the vast majority of other results differed dramatically. The inflated background increased false positives and reduced the reliability of the enrichment findings

Figure 1. Comparison of the pathway enrichment analysis results using the two different backgrounds. Enrichment results clearly illustrate the impact of background selection on the number and significance of pathways detected.

At first glance, you might think: “Well, the first pathway is the same. So, if a pathway is truly relevant, it will appear at the top no matter what background is used.” Unfortunately, this isn’t the case. The overlap at the top is coincidental and doesn’t address the larger issue: inappropriate background selection can dramatically distort statistical results.

To further illustrate, consider a simplified example. We analyzed 36,000 measured genes, of which 10% (3,600) were differentially expressed. Suppose 100 genes are associated with a single arbitrary pathway, and 12 of these appear in our differentially expressed list.

When we use the total list of measured genes as the background, the pathway’s p-value is 0.19—indicating no significant enrichment. But analyzing the exact same data against the larger NCBI reference set (~52,000 genes) drops the p-value to 0.02, falsely indicating the pathway as significant.

Table 1: Comparison of enrichment results on a single pathway using different backgrounds. Using the larger NCBI reference set (background gene set) inflates statistical significance, illustrating how inappropriate background selection can lead to false positives.

Proper background selection and thoughtful management of large lists of differentially expressed genes (DEGs) are crucial for meaningful and reliable pathway analysis results. The following best practices can help you avoid common pitfalls and improve the biological relevance of your findings.

• Use  all the measured genes or proteins as your analysis background. This ensures statistical validity and reduces false positives in enrichment analysis.
• Avoid using arbitrary or overly broad gene sets from external databases as your enrichment background. These can inflate pathway significance and skew results.
• Refine your list of DEGs by applying more stringent thresholds for p-values and fold changes when dealing with large DEG sets. This helps focus on the most biologically relevant genes.
• Re-run pathway analysis with updated gene lists and proper background selection to improve reliability and biological insight.

On Advaita’s cloud based multi-omics analysis platform, iPathwayGuide™, we require users to submit either their complete list of measured genes or explicitly specify the background set. This approach aligns with best practices for pathway and enrichment analysis and ensures the statistical validity of your results. By doing this, iPathwayGuide™ prevents the common pitfall of using arbitrary or inflated background gene pools from external databases, which can lead to artificially significant pathways and unreliable conclusions.

If you’re using an omics analysis tool that asks only for your list of significant genes, ask yourself: How can it calculate enrichment without knowing the full measured pool of genes or proteins from your experiment? Without this information, the tool is forced to assume an arbitrary background, often all the genes in a reference database. This can dramatically skew p-values and lead to inaccurate biological interpretations.

• The background gene or protein set used in pathway and other enrichment analyses directly impacts the statistical significance of your results.
• Always use all the measured genes or proteins from your experiment as the background to avoid inflated p-values and false positives.
• Arbitrary backgrounds from external databases can distort enrichment results and lead to misleading biological interpretations.
• iPathwayGuide™ enforces best practices by requiring explicit background specification, ensuring more reliable and scientifically valid findings.

• Chapter 24 in Statistics and Data Analysis for Microarrays Using R and Bioconductor, Second Edition (Chapman & Hall/CRC Mathematical and Computational Biology)
• Background Gene List for Enrichment Analysis, Metascape Blog (Aug 3 2019)
• Avoiding the Pitfalls of Gene Set Enrichment Analysis with SetRank, SetRank Blog (Sept 10 2020)
• Gene Set Analysis: A Step-By-Step Guide, Nature Protocols (Oct 2016)
• Bioinformatics Enrichment Tools: Paths Toward the Comprehensive Functional Analysis of Large Gene Lists, Nucleic Acids Research (Jan 2009)
• Gene Set Enrichment Analysis: Performance Evaluation and Usage Guidelines, BMC Bioinformatics (Feb 2021)