Choosing the background gene list for enrichment analysis
The background is the set of genes your significant genes are compared against. Getting it wrong is one of the most common causes of misleading enrichment results.
What the background does
Over-representation analysis asks whether a pathway holds more of your significant genes than expected. "Expected" needs a reference: out of which genes could yours have been drawn? That reference is the background, sometimes called the gene universe.
The common mistake
Many web tools take only a pasted list of significant genes. With nothing else to go on, they compare it against every annotated gene in the genome.
The trouble is that your experiment could never have found most of those genes. A liver sample expresses perhaps 12,000 to 15,000 genes. Genes that are silent in liver, such as olfactory receptors, cannot be differentially expressed there, yet they count toward the genome-wide background.
The result is that pathways made of genes expressed in your tissue look enriched simply because they are expressed. A liver experiment tends to return liver pathways whatever the treatment was.
The right background for RNA-seq
Use the genes that were actually tested: every gene in your results table that received a p-value. This is the set from which your significant genes were drawn.
- In DESeq2, that means genes with a non-NA
padj. - In edgeR or limma, it means the genes left after expression filtering.
How to supply it
- Tools that take a results table can work out the background themselves. ExpressionBench DE Results does this: it uses every gene in your uploaded table that is annotated in the chosen gene-set library, and shows the number on screen.
- Tools that take a gene list usually have an optional field for a custom background. Paste your tested genes there.
How much difference it makes
It depends on the tissue and the pathway, but the effect is systematic: a genome-wide background makes p-values look smaller than they should for anything related to the tissue's normal function. Results that survive a correct background are the ones worth following up.
What to report
State the background in your methods, with its size. For example: "The background was the 14,212 genes tested for differential expression." A reader, or a reviewer, can then judge the result.
Try it on your own results. The DE Results tool on ExpressionBench turns one results table into a volcano plot, a gene table and pathway enrichment. It is free, and your data stays in your browser.
Open the tool