Enrichment analysis
Material
- MSigDB
clusterProfiler
vignette- Revigo
- Signaling Pathway Impact Analysis (SPIA)
- Original paper on GSEA
- STRING for protein-protein interactions
- GO figure! for plotting GO terms and the associated paper
Exercises
Load the following packages:
library(clusterProfiler)
library(enrichplot)
If the FindMarkers
or FindAllMarkers
functions were used,
we obtained a table listing only the significant genes,
but we don’t have any information of fold change for the non-significant
genes. Therefore, we can use the over-representation analysis
which is a threshold-based method.
Using our list of significant genes, we can test
if any gene set is over-represented among significant genes or not using a test
similar to a Fisher test to compare differences in proportions.
The clusterProfiler
package provides functions for over-representation
analysis of Gene Ontology gene sets (among other functions, including functions for actual GSEA) or KEGG gene sets.
Genes can be labeled using different types of labels, eg symbol, Ensembl ID, Entrez ID. To list the allowed label types use:
BiocManager::install("org.Hs.eg.db", update = FALSE)
library(org.Hs.eg.db)
AnnotationDbi::keytypes(org.Hs.eg.db)
Let’s select a set of genes that are downregulated in the tumor cells compared to normal:
tum_down <- subset(tum_vs_norm,
tum_vs_norm$avg_log2FC < -1 &
tum_vs_norm$p_val_adj < 0.05)
tum_down_genes <- rownames(tum_down)
We can do a Gene Ontology term over-representation analysis based on this set of genes. Make sure you check out the help of this function to understand its arguments:
?enrichGO
tum_vs_norm_go <- clusterProfiler::enrichGO(tum_down_genes,
"org.Hs.eg.db",
keyType = "SYMBOL",
ont = "BP",
minGSSize = 50)
The results are stored in the @result
slot:
View(tum_vs_norm_go@result)
Some GO terms seem redundant because they contain many of the same genes, which is a characteristic of Gene Ontology gene sets. We can simplify this list by removing redundant gene sets:
enr_go <- clusterProfiler::simplify(tum_vs_norm_go)
View(enr_go@result)
We can quite easily generate a plot called an enrichment map with the enrichplot
package:
enrichplot::emapplot(enrichplot::pairwise_termsim(enr_go),
showCategory = 30, cex_label_category = 0.5)

Instead of testing for Gene Ontology terms, we can also test for other gene set collections. For example the Hallmark collection from MSigDB:
gmt <- msigdbr::msigdbr(species = "human", category = "H")
We can use the function enricher
to test for over-representation of any set of genes of the Hallmark collection. We have to include the “universe”, i.e. the full list of background, non significant genes, against which to test for differences in proportions:
tum_vs_norm_enrich <- clusterProfiler::enricher(gene = tum_down_genes,
universe = rownames(proB),
pAdjustMethod = "BH",
pvalueCutoff = 0.05,
qvalueCutoff = 0.05,
TERM2GENE = gmt[,c("gs_name", "gene_symbol")])
When using the genes down-regulated in tumor, among the over-represented Hallmark gene sets, we have HALLMARK_G2M_CHECKPOINT
, which includes genes involved in the G2/M checkpoint in the progression through the cell division cycle.
View(tum_vs_norm_enrich@result[which(tum_vs_norm_enrich@result$p.adjust<0.05),])
Clear environment
Clear your environment:
rm(list = ls())
gc()
.rs.restartR()