Biological Interpretation of Microarray Data
Helen Lockstone
DTC Bioinformatics Course
9th February 2010
Overview
• Interpreting microarray results– Gene lists to biological knowledge
• The Gene Ontology Consortium– Defined terms to describe gene function
• Functional analysis tools– Methods– DAVID/GSEA
Microarray Pipeline
Design and perform experiment
Process and normalise data
Statistical analysis
Differentially expressed genes
Biological interpretation
Biological Interpretation
• An obvious way to gain biological insight is to assess the differentially expressed genes in terms of their known function(s)
• Required an automated and objective (statistical) approach
• Functional profiling or pathway analysis
Early functional analyses
• Manually annotate list of differentially expressed (DE) genes
• Extremely time-consuming, not systematic, user-dependent
• Group together genes with similar function• Conclude functional categories with most DE
genes important in disease/condition under study• BUT may not be the right conclusion
GO and functional analysis
Immune response
Metabolism
Transcription
Energy production
Neurotransmission
Protein transport
Functional category Number of sig genesImmune response 40Metabolism 20Transcription 20Energy production 10Neurotransmission 5Protein transport 5TOTAL 100
Immune response category contains 40% of all significant genes - by far the largest category.
Reasonable to conclude that immune response may be important in the condition being studied?
However ….
• What if 40% of the genes on the array were involved in immune response?
• Only detected as many significant immune response genes as expected by chance
• Need to consider not only the number of significant genes for each category, but also total number on the array
Same example, relative to array
Functional category
Number of genes on array
Actual number of significant genes
Expected number of significant genes
Immune response 8000 40 40Metabolism 4000 20 20Transcription 2000 20 10Energy production 4000 10 20Neurotransmission 200 5 1Protein transport 1800 5 9
ALL 20000 100
Expected number of significant genes for category X = (num sig genes ÷ total genes on array)*(num genes in category X on array)
Same example, relative to array
• Now, transcription and neurotransmission categories appear more interesting as many more significant genes were observed than expected by chance
• Largest categories are not necessarily the most interesting!
Functional category
Number of genes on array
Actual number of significant genes
Expected number of significant genes
Immune response 8000 40 40Metabolism 4000 20 20Transcription 2000 20 10Energy production 4000 10 20Neurotransmission 200 5 1Protein transport 1800 5 9
ALL 20000 100
Major bioinformatic developments
• Requires annotating entire set of genes
• The Gene Ontology Consortium (www.geneontology.org)
• Automated, statistical approaches for annotating gene lists and performing functional profiling
The Gene Ontology Consortium
GO Consortium
• Developed three structured and controlled vocabularies (ontologies) that describe gene products in terms of their associated biological processes, cellular components and molecular functions in a species-independent manner
• Has become a major resource for microarray data interpretation
The Gene Ontology
• Molecular Function: basic activity or task
• Biological Process: broad objective or goal
• Cellular Component: location or complex
The Gene Ontology
• Molecular Function: basic activity or task
– e.g. catalytic activity, calcium ion binding
• Biological Process: broad objective or goal
– e.g. signal transduction, immune response
• Cellular Component: location or complex
– e.g. nucleus, mitochondrion
GO Structure
• Hierarchical tree• Annotated with most
specific annotation, forming path to top of tree
• Genes annotated with all relevant terms
• Annotations based on published studies and also electronic inferences
GO Terms
• GO ID: GO:0007268
• GO term: synaptic transmission
• Ontology: biological process
• Definition: The process of communication from a neuron to a target (neuron, muscle, or secretory cell) across a synapse
Graphical view
http://www.ncbi.nlm.nih.gov/sites/entrez
Functional Profiling Tools
Functional profiling tools
Identify GO categories with significantly more DE genes than expected by chance (i.e. over-represented among DE genes relative to
representation on array as a whole)
Correct for testing multiple GO categories
Hypergeometric Distribution or Fisher’s Exact Test
Khatri and Draghici. Ontological analysis of gene expression data: current tools, limitations, and open problems. Bioinformatics (2005) 21(18):3587-95
Functional profiling tools
Functional profiling tools
• Freely-available stand-alone/web-based tools– User-friendly graphical interface and simple to use– Extensive documentation, plus tutorials/technical support
• Reduces a large number of DE genes to a smaller number of significantly enriched GO categories – more easily interpreted in biological context
• Considering sets of genes increases power – individual genes could be false positives but a set of functionally
related genes all showing significant changes is more robust
DAVID Results
Advantages
• Increasingly support data (probe IDs) from different microarray platforms
• Accept various probe/gene identifiers
• Web-based tools automatically retrieve most up-to-date GO annotations
• Most automatically map from probe IDs to a gene ID - multiple significant probes for one gene could otherwise skew results
Further considerations
• Reference list must be appropriate for accurate statistical analysis
• Up/down regulated genes can be submitted separately or as a combined list
• Unannotated genes cannot be used in the analysis; gene ontology evolving; well-studied systems over-represented
Gene set enrichment analysis
• Majority of tools based on idea of identifying GO categories significantly enriched in list of differentially expressed genes
• Requires some threshold to define genes as ‘significant’
• Recent tool called GSEA takes a different approach by considering all assayed genes
GSEA: Key Features
• Ranks all genes on array based on their differential expression
• Identifies gene sets whose member genes are clustered either towards top or bottom of the ranked list (i.e. up- or down regulated)
• Enrichment score calculated for each category • Permutation test to identify significantly enriched
categories• Extensive gene sets provided via MolSig DB – GO,
chromosome location, KEGG pathways, transcription factor or microRNA target genes
GSEA
• Each gene category tested by traversing ranked list
• Enrichment score starts at 0, weighted increment when a member gene encountered, weighted decrement otherwise
• Enrichment score – point where most different from zero
Most significantly up-regulated genes
Unchanged genes
Most significantly down-regulated genes
Disease Control
GSEA algorithm
Null distribution of enrichment scores
Actual ES
GSEA: Permutation Test
• Randomise data (groups), rank genes again and repeat test 1000 times
• Null distribution of 1000 ES for geneset
• FDR q-value computed – corrected for gene set size and testing multiple gene sets
Biological Interpretation
• Due to GO hierarchy, several related categories may contain a subset of genes that is driving the significant enrichment score so will all be significant
• Interpretation still requires substantial work– search literature and public databases – likely functional consequences of the changes– are the genes identified as significant within each GO
category up- or down-regulated?– genes within a category can have opposite effects e.g.
apoptosis would include genes that induce or repress apoptosis
Biological Interpretation
• Too many categories found significant– Size filter – More stringent significance threshold– Related categories (redundancy)
• No significant categories– Relax significance level slightly – e.g. 0.25 recommended by GSEA as exploratory analysis
• No significant genes– GSEA most suitable
Commercial Tool Suites
• Ingenuity Pathway Analysis (Ingenuity Systems, CA)– Developed own extensive ontology over past 10 years – Includes gene interactions, disease/drug information– PhD-level curators mining the literature– Used by many pharmaceutical companies
For more information
• Gene Ontology: http://www.geneontology.org • Affymetrix: http://www.affymetrix.com • DAVID: http://david.abcc.ncifcrf.gov• GSEA: http://www.broad.mit.edu/gsea/ • Ingenuity:
http://www.ingenuity.com/products/pathways_analysis.html
• NCBI: http://www.ncbi.nlm.nih.gov/
Top Related