Gene set analyses of genomic datasets Andreas Schlicker Jelle ten Hoeve Lodewyk Wessels

Scenario You have a gene expression dataset containing data from normal colon and adenoma samples. - Which pathways are differentially regulated between normal and CRC samples? -Do products of significantly differently expressed genes have specific functions (Gene Ontology)? -Is there a significant overlap with published expression signatures (mutations, response to treatment,...)?

Overview Mapping probe sets to functional annotation Hypergeometric test (Fisher’s exact test) Gene Set Enrichment Analysis Global test

Mapping probe sets to functional annotation

Examples of functional annotation Pathway databases (e.g. KEGG, Pathway Interaction Database, ConsensusPathDB, Functional categories (e.g. Gene Ontology, FunCat) Enzyme Commission numbers, disease associations, protein domains, … Published gene signatures

Example KEGG pathway

Gene Ontology Collection of three separate ontologies: biological process, molecular function, cellular component Organized in a graph structure, i.e. each term (concept, category) can have several parents

Gene Ontology (II)

Gene Ontology (III) Annotations with GO terms are assigned an evidence code: G protein alpha subunit; GO: activation of phospholipase C …; ISS Different categories of evidence codes: experimental, computational, Author/Curator statement, fully automatic (IEA) Details at

The true path rule If a gene product is annotated with term A, all annotations with ancestors of A must also be valid. Gene product annotated with this term  It can also be annotated with the term‘s ancestors Different gene products are usually not annotated on the same level of the hierarchy

Hands on Time

The hypergeometric test / Fisher’s exact test

Basics Enrichment test Analysis steps: 1.Single gene test (e.g. t-test for finding differentially expressed genes) 2.Do list (step 1) and gene sets overlap significantly? diff. Expressednot diff. expressed in gene set not in gene set

Example Microarray: 20000, MAPK: 100, diff. expressed: 200  Fisher‘s exact test p = 0.26 diff. Expressed not diff. expressed total MAPK not MAPK total

Example Microarray: 20000, MAPK: 100, diff. expressed: 200  Fisher‘s exact test p = diff. Expressed not diff. expressed total MAPK not MAPK total

Another Example Consider having data on treatment response and gene mutation for samples in a dataset ! Choose threshold for resistance/sensitivity ResistantSensitivetotal Mutated WT total

Problem with this approach Null hypothesis: Genes in the gene set are randomly drawn  Significant result means that genes in the gene set are more alike than random genes Problem: Gene set has been selected such that the genes have something in common  False positives

Hands on Time

PAGE: Parametric Analysis of Gene Set Enrichment

Basics For each gene set and each sample: –How different is the mean expression of all genes in a gene set from the overall mean expression? Applied to full expression matrix –No need for selecting interesting genes (based on e.g. t-test)

Basics

Problem with this approach What happens if one part of the pathway is up-regulated and the another part is down-regulated?

Hands on Time

The global test

Basics Group test Can the genes in the gene set predict the response? What is needed? –Clinical variablee.g. normal vs. CRC –Gene expressione.g. GSE8671 –Gene setse.g. KEGG pathways

Interpretation Interpretation of significant test result (w.r.t. genes): –Gene set is associated with clinical variable –“On average“ the genes in the set are associated with the clinical variable –Not every gene needs to be associated

Interpretation

Interpretation of significant test result (w.r.t. samples): –Expression profile in the gene set differs for different values of the clinical variable –Samples with similar value (clinical variable) have relatively similar expression profiles

Interpretation