Persistent Systems Pvt. Ltd. http://www.persistent.co.in

1
Gene Expression Analysis Using Microarrays

• Dr Mushtaq Ahmed
• Technology Incubation Division
• Persistent Systems Private Ltd
• Pune

2
Topics

1. Introduction
2. Data Storage and Exchange Standards
3. Analysis (Clustering)
4. Conclusion and References

3
1. Introduction

• Structure Activity Relationship
• Structural vs. Functional Genomics
• Principals of Microarray Experiment
• Applications

4
Structure Activity Relationship
GENES (finite)
EXPERIMENTAL SETUP
Functional Genomics OR Confirmation Work
Structural Genomics OR Prediction Work
FUNCTIONS (infinite)
PROTEINS
5
SourceYale Bioinformatics
6
Principles of a Microarray ExperimentHybridizati
on

• Environment ? Functions ? Proteins ? mRNA ? cDNA
• Different incubations of cells results in up or
  down regulation of different sets of genes.
• Microarray provides a medium for matching known
  and unknown DNA samples based on base-pairing
  rules and automating the process of identifying
  the unknowns
• Set of expressed genes (at mRNA stage) isolated
  and identified using hybridization on a
  microarray chip

7
HTS Using Hybridization
Microarray Chip
Target cDNA (variables to be detected)
Probe oligos/cDNA (gene templates)

Samples
Hybridization
Analysis of outcome
Pathways
Functional Annotation
Targets/Leads
Disease Class.
Physiological states
8
Timeline for drug discovery
Discovery (5 yrs) 5000 Gene expression
study Pre-Clinical (1 yr) 50 Clinical (6
yrs) 5 Review (2 yrs) 1 Marketed
9
2. Data Storage and Exchange Standards

• Raw and Processed Data
• Conceptual View of Database
• Example of ArrayExpress
• Issues
• Standardization for Exchange

10
Raw data images

• Red (Cy5) dot
• overexpressed or up-regulated
• Green (Cy3) dot
• underexpressed or down-regulated
• Yellow dot
• equally expressed
• Intensity - absolute level
• red/green - ratio of expression
• 2 - 2x overexpressed
• 0.5 - 2x underexpressed
• log2( red/green ) - log ratio
• 1 2x overexpressed
• -1 2x underexpressed

cDNA plotted microarray
11
Microarray Expression Value Representation
expression value types
composite spots
primary measurements
derived values
primary spots
composite images e.g., green/red ratios
primary images
Source MGED
12
Gene expression database a conceptual view
Samples
Gene expression matrix
Genes
Gene expression levels
13
(No Transcript)
14
DAG Representation of Biomaterials
Source MGED
15
ArrayExpress (MGED) Design
Source MGED
16
ArrayExpress (MGED) Architecture
application server
Web server
MAML data
ArrayExpress
data warehouse
data submission Curation database
image server?
Curation pipeline
Source MGED
17
Issues in Storage

• Size of Data
• Experiments
• 100 000 genes, 320 cell types
• 2000 compounds, 3 time points, 2 concentrations,
  2 replicates
• Data
• 8 x 1011 data-points
• 1 x 1015 1 petaB of data
• Others
• Raw data are images
• lack of standard measurement units for gene
  expression
• lack of standards for sample annotation

18
Standardization

• MIAME (Minimum Info About a Microarray Expt)
• Experimental design, Array design
• Samples, Hybridisations
• Measurements, Controls
• OMG-LSR-DFT
• Life Sciences Research, Domain Task Force Gene
  Expression RFP
• EBI (MAML), Rosetta (GEML), NetGenics
  submitters
• Proposed MAGEML (MAML GEML)
• Annotations data data stored as a set of
  external 2D matrices
• Data format independent of particular scanner or
  image analysis software
• Sample and treatment can be represented as a
  Directed Acyclic Graphs
• Concept of composite images and composite spots

19
3. Data Analysis (Clustering)

• Normalization
• Hierarchical Clustering
• Divisive Clustering
• Other Methods
• Visual Tools

20
Normalization

• Assumption
• Average expression ratio 1
• Amount of mRNA from both the sample is same
• Total Intensity
• Calculate a factor to rescale intensities of all
  te genes so that
• total Cy3 total Cy5
• Regression Techniques
• Adjust the intensities so that
• Slope of scatter plot of Cy3 vs Cy5 1
• Using ratio statistics
• Based on housekeeping genes expression a
  probability density ratio is developed which is
  used for normalization

21
(No Transcript)
22
Clustering

• Hierarchical
• Single, Complete and Average Linkage
• Divisive
• K-means
• Self Organizing Maps (SOM)
• Others
• Principal Component Analysis (PCA)
• Supervised Methods

23
Hierarchical clustering

• Distance metrics or Similarity Measures
• Euclidian, Pearson, distance of slopes etc..
• Cost functions
• Single Linkage
• Min distance of any two members (one from each of
  the two clusters)
• Complete Linkage
• Max distance of any two members (one from each of
  the two clusters)
• Average Linkage
• UPGMA
• WPGMA
• Within Groups
• Wards Method
• Join which produces smallest possible error in
  some of squared errors

24
(No Transcript)
25
Divisive clustering

• K-means
• k random (or specified) points used to create
  clusters, average vectors for the clusters then
  used iteratively
• Knowledge of probable no of clusters (k) needed
• Used in combination with PCA and hierarchical
  clustering
• Self Organizing maps
• User defined geometric configurations as
  partitions
• Random vectors generated for each partition and
  TRAINED till convergence (ANN based)
• Visualization Methods
• Helps in cluster visualization
• Scatter Plot, Web plot, histogram
• May help in clustering itself
• E.g., SuperGrouper utility of MaxdView

26
(No Transcript)
27
Other Clustering Methods

• PCA (Principal Component Analysis)
• Also called SVD (Singular Value Decomposition)
• Reduces dimensionality of gene expression space
• Finds best view that helps separate data into
  groups
• Supervised Methods
• SVM (Support Vector Machine)
• Previous knowledge of which genes expected to
  cluster is used for training
• Binary classifier uses feature space and
  kernel function to define a optimal
  hyperplane
• Also used for classification of samples-
  expression fingerprinting for disease
  classification

28
(No Transcript)
29
4. Conclusion and References

• Microarrays makes HTS with hybridization possible
• No single standard unit for measuring expression
  levels
• Handling and interpretation not yet exact
• Assumptions Elements in cluster must share some
  commonality
• Classification depends on method used for
  clustering, normalization, distance function
• No correct way of classification, biological
  understanding is the ultimate guide
• Provides extension to existing knowledge (e.g.,
  classifying a novel gene into a known pathway)

30
Software

• Databases
• Public repositories
• GEO (NCBI), GeneX (NCGR), ArrayExpress (EBI)
• In-house databases
• Stanford, MIT, University of Pennsylvania,
• Organism specific databases
• Mouse Genome Informatics Database
• Proprietary databases
• Gene Logic, NCI, Synergy (NetGenics), Genomics
  Knowledge Platform (Incyte)
• Analysis Tools
• Public Domain
• maxdView (University of Manchester)
• CyberT , RCuster interfaces of GeneX
• Proprietary
• Spotfire, Xpression NTI (Informaxinc)

31
References

• Microarray Gene Expression Database Group
• http//www.mged.org
• National Center for Genomic Research
• http//genex.ncgr.org
• University of Manchester , Bioinformatics Group
• http//bioinf.man.ac.uk/microarray/resources.html
• Nature Reviews Genetics
• http//www.nature.com/nrg/