5 Overview of enrichment analysis

5.1 Terminology

5.1.1 Gene sets and pathway

A gene set is an unordered collection of genes that are functionally related. A pathway can be interpreted as a gene set by ignoring functional relationships among genes.

5.1.2 Gene Ontology (GO)

Gene Ontology defines concepts/classes used to describe gene function, and relationships between these concepts. It classifies functions along three aspects:

  • MF: Molecular Function
    • molecular activities of gene products
  • CC: Cellular Component
    • where gene products are active
  • BP: Biological Process
    • pathways and larger processes made up of the activities of multiple gene products

GO terms are organized in a directed acyclic graph, where edges between terms represent parent-child relationship.

5.1.3 Kyoto Encyclopedia of Genes and Genomes (KEGG)

KEGG is a collection of manually drawn pathway maps representing molecular interaction and reaction networks. These pathways cover a wide range of biochemical processes that can be divided into 7 broad categories:

  1. Metabolism
  2. Genetic information processing
  3. Environmental information processing
  4. Cellular processes
  5. Organismal systems
  6. Human diseases
  7. Drug development.

5.1.4 Other gene sets

GO and KEGG are the most frequently used for functional analysis. They are typically the first choice because of their long-standing curation and availability for a wide range of species.

Other gene sets include but are not limited to Disease Ontology (DO), Disease Gene Network (DisGeNET), wikiPathways, Molecular Signatures Database (MSigDb).

5.2 Over Representation Analysis

Over Representation Analysis (ORA) (Boyle et al. 2004) is a widely used approach to determine whether known biological functions or processes are over-represented (= enriched) in an experimentally-derived gene list, e.g. a list of differentially expressed genes (DEGs).

The p-value can be calculated by hypergeometric distribution.

\(p = 1 - \displaystyle\sum_{i = 0}^{k-1}\frac{{M \choose i}{{N-M} \choose {n-i}}} {{N \choose n}}\)

In this equation, N is the total number of genes in the background distribution, M is the number of genes within that distribution that are annotated (either directly or indirectly) to the gene set of interest, n is the size of the list of genes of interest and k is the number of genes within that list which are annotated to the gene set. The background distribution by default is all the genes that have annotation. P-values should be adjusted for multiple comparison.

Example: Suppose we have 17,980 genes detected in a Microarray study and 57 genes were differentially expressed. Among the differentially expressed genes, 28 are annotated to a gene set1.

d <- data.frame(gene.not.interest=c(2613, 15310), gene.in.interest=c(28, 29))
row.names(d) <- c("In_category", "not_in_category")
##                 gene.not.interest gene.in.interest
## In_category                  2613               28
## not_in_category             15310               29

Whether the overlap(s) of 25 genes are significantly over represented in the gene set can be assessed using a hypergeometric distribution. This corresponds to a one-sided version of Fisher’s exact test.

fisher.test(d, alternative = "greater")
##  Fisher's Exact Test for Count Data
## data:  d
## p-value = 1
## alternative hypothesis: true odds ratio is greater than 1
## 95 percent confidence interval:
##  0.110242      Inf
## sample estimates:
## odds ratio 
##  0.1767937

5.3 Gene Set Enrichment Analysis

A common approach to analyzing gene expression profiles is identifying differentially expressed genes that are deemed interesting. The ORA enrichment analysis is based on these differentially expressed genes. This approach will find genes where the difference is large and will fail where the difference is small, but evidenced in coordinated way in a set of related genes. Gene Set Enrichment Analysis (GSEA)(Subramanian et al. 2005) directly addresses this limitation. All genes can be used in GSEA; GSEA aggregates the per gene statistics across genes within a gene set, therefore making it possible to detect situations where all genes in a predefined set change in a small but coordinated way. This is important since it is likely that many relevant phenotypic differences are manifested by small but consistent changes in a set of genes.

Genes are ranked based on their phenotypes. Given apriori defined set of gene S (e.g., genes sharing the same DO category), the goal of GSEA is to determine whether the members of S are randomly distributed throughout the ranked gene list (L) or primarily found at the top or bottom.

There are three key elements of the GSEA method:

  • Calculation of an Enrichment Score.
    • The enrichment score (ES) represents the degree to which a set S is over-represented at the top or bottom of the ranked list L. The score is calculated by walking down the list L, increasing a running-sum statistic when we encounter a gene in S and decreasing when it is not encountered. The magnitude of the increment depends on the gene statistics (e.g., correlation of the gene with phenotype). The ES is the maximum deviation from zero encountered in the random walk; it corresponds to a weighted Kolmogorov-Smirnov(KS)-like statistic (Subramanian et al. 2005).
  • Esimation of Significance Level of ES.
    • The p-value of the ES is calculated using a permutation test. Specifically, we permute the gene labels of the gene list L and recompute the ES of the gene set for the permutated data, which generate a null distribution for the ES. The p-value of the observed ES is then calculated relative to this null distribution.
  • Adjustment for Multiple Hypothesis Testing.
    • When the entire gene sets are evaluated, the estimated significance level is adjusted to account for multiple hypothesis testing and also q-values are calculated for FDR control.

We implemented the GSEA algorithm proposed by Subramanian (Subramanian et al. 2005). Alexey Sergushichev implemented an algorithm for fast GSEA calculation in the fgsea (Korotkevich, Sukhov, and Sergushichev 2019) package. In our packages (clusterProfiler, DOSE, meshes and ReactomePA), users can use the GSEA algorithm implemented in DOSE or fgsea by specifying the parameter by="DOSE" or by="fgsea". By default, the fgsea method will be used since it is much more faster.

5.4 Leading edge analysis and core enriched genes

Leading edge analysis reports Tags to indicate the percentage of genes contributing to the enrichment score, List to indicate where in the list the enrichment score is attained and Signal for enrichment signal strength.

It would also be very interesting to get the core enriched genes that contribute to the enrichment. Our packages (clusterProfiler, DOSE, meshes and ReactomePA) support leading edge analysis and report core enriched genes in GSEA analysis.