Enrichment analysis

Material

Exercises

Load the following packages:

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)

About OrgDb

For other organisms, you can find available OrgDbs at bioconductor

Let’s select a set of genes that are downregulated in the tumor cells compared to normal:

tum_down  <- subset(limma_de,
                    limma_de$logFC < -1 
                      & limma_de$adj.P.Val <  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)
ID Description GeneRatio BgRatio pvalue p.adjust qvalue
GO:0007059 GO:0007059 chromosome segregation 105/809 424/18870 0 0 0
GO:0098813 GO:0098813 nuclear chromosome segregation 79/809 312/18870 0 0 0
GO:0000070 GO:0000070 mitotic sister chromatid segregation 62/809 184/18870 0 0 0
GO:0000819 GO:0000819 sister chromatid segregation 67/809 225/18870 0 0 0
GO:0140014 GO:0140014 mitotic nuclear division 71/809 274/18870 0 0 0
GO:0000280 GO:0000280 nuclear division 88/809 441/18870 0 0 0
The columns GeneRatio and BgRatio

The columns GeneRatio and BgRatio that are in the enrichResult object represent the numbers that are used as input for the Fisher’s exact test.

The two numbers (M/N) in the GeneRatio column are:

  • M: Number of genes of interest (in our case tum_down_genes) that are in the GO set
  • N: Number of genes of interest with any GO annoation.

The two numbers (k/n) in the BgRatio column are:

  • k: Number of genes in the universe that are in the GO set
  • n: Number of genes in the universe with any GO annoation

A low p-value resulting from the Fisher’s exact means that M/N is signficantly greater than k/n. 

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)
ID Description GeneRatio BgRatio pvalue p.adjust qvalue
GO:0007059 GO:0007059 chromosome segregation 105/809 424/18870 0 0 0
GO:0098813 GO:0098813 nuclear chromosome segregation 79/809 312/18870 0 0 0
GO:0000070 GO:0000070 mitotic sister chromatid segregation 62/809 184/18870 0 0 0
GO:0000280 GO:0000280 nuclear division 88/809 441/18870 0 0 0
GO:0044772 GO:0044772 mitotic cell cycle phase transition 77/809 470/18870 0 0 0
GO:0051983 GO:0051983 regulation of chromosome segregation 40/809 131/18870 0 0 0

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.params = list(category_label = 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[tum_vs_norm_enrich@result$p.adjust < 0.05,])
ID Description GeneRatio BgRatio pvalue p.adjust qvalue
HALLMARK_E2F_TARGETS HALLMARK_E2F_TARGETS HALLMARK_E2F_TARGETS 80/344 195/3866 0.0000000 0.0000000 0.0000000
HALLMARK_G2M_CHECKPOINT HALLMARK_G2M_CHECKPOINT HALLMARK_G2M_CHECKPOINT 67/344 188/3866 0.0000000 0.0000000 0.0000000
HALLMARK_MITOTIC_SPINDLE HALLMARK_MITOTIC_SPINDLE HALLMARK_MITOTIC_SPINDLE 48/344 197/3866 0.0000000 0.0000000 0.0000000
HALLMARK_MYC_TARGETS_V1 HALLMARK_MYC_TARGETS_V1 HALLMARK_MYC_TARGETS_V1 33/344 194/3866 0.0001564 0.0017991 0.0016468
HALLMARK_ESTROGEN_RESPONSE_LATE HALLMARK_ESTROGEN_RESPONSE_LATE HALLMARK_ESTROGEN_RESPONSE_LATE 28/344 166/3866 0.0005747 0.0052875 0.0048398

Clear environment

Clear your environment:

rm(list = ls())
gc()
.rs.restartR()