Applications in Bioinformatics Identifying Differential Gene Expression

1
Applications in Bioinformatics Identifying
Differential Gene Expression
2
Acknowledgments

• Collaborators
• John Oleskiewicz, GVSU
• Dr. Kyle Furge, Van Andel Research Institute
• Research Support
• Science Math Summer Research Award
• Travel Support
• For presentation at IEEE Bioinformatics
  Conference, Stanford University, Palo Alto,
  August 2002.

3
First Thoughts

• Perhaps bioinformatics--the shotgun marriage
  between biology, mathematics, computer science,
  and engineering--is like an elephant that
  occupies a large chair in the scientific living
  room ... There are probably many biologists who
  feel that a major product of this bioinformatics
  elephant is large piles of waste material."
• -- Sylvia J. Spengler Science Feb.
  18, 2000.

4
Bioinformatics in Context

• Evolution of the Scientific Method
• Empirical science observation
• Theoretical science modeling
• Computational science simulation
• in vivo ? in vitro ? in silica
• Computational Biology analytical computations
• Bioinformatics information management, data
  mining

5
Overview

• Gene expression
• DNA ? transcription ? mRNA ? translation ?
  protein
• Genomic gains/losses
• Clinical prognosis
• Differences in gene expression level
• Regulatory gene defects
• DNA mutations
• Single Nucleotide Polymorphisms

6
Microarray Technology

• Lithographic technique
• cDNA microchip
• Robotic application
• Oligonucleotides
• Multiple patient samples

7
Microarray Implementation
DNA clones
mRNA
Control
Sample
Reverse transcription (cDNA)
PCR amplification
Fluorescent labeling
Purification
Hybridization
8
Microarray Scanning
Laser excitation
C C D detection
Computer analysis
Scanned image
9
Microarray Analysis

• Rows ? genes
• Columns ? samples
• Identifies
• Expression patterns
• Expression levels

10
Project Goal GAT

• Problem
• Hundreds of samples ? thousands of genes
• Goal
• Improve existing analysis tools
• Develop parallel version
• Model
• CIT (Cluster Identification Tool)
• Method
• Reverse engineering
• Consultation
• Current papers

11
Experimental Process

• Objectives
• Group experimental samples by user-defined
  parameters
• Calculate statistical metrics on gene expression
  data (t-stat, discrimination score,
  p-value, false discovery rate)
• Employ statistical discrimination and permutation
  analysis to identify genes that differentiate
  between the groups
• Generate lists of discriminating genes sorted by
  significance

12
Discriminant Analysis

• Finding significant discriminating genes
• Identify differentially expressed genes
  students t-statistic
• Measure significance of difference
• compare sample group t-stat to a random
  distribution
• Calculate d-score, P-value
• Estimate FDR

13
Identifying DifferentialGene Expression
Group1
Group2
Unsorted
GAT
14
Performance Comparison

• Machine 1.5 GHz P4 w/256 MB
• Microarray 4550 genes
• Samples 8 melanoma cell lines, 52 others
• Analysis Package
• VARI CIT 9.4
• Stanford SAM
• Alternative statistical approach
• GAT Linux 2.4 w/gcc
• GAT Windows 2000 Pro w/Visual C

15
Performance Execution Time
16
Distributed Mode GAT
10.152.31.1 FreeBSD listening port 3307 processed
1101
Server 1
Perl Client Connecting... 10.152.31.1 10.152.31.2
10.152.31.3
10.152.31.2 Linux listening port 3308 processed
700
Server 2
GAT Client
10.152.31.3 Windows 2000 listening port
3307 processed 110
Server N
17
Experimental Parameters

• Machines 900 MHz PIII w/256 MB
• Microarray 4550 genes
• Samples 8 melanoma cell lines, 52 others
• Analysis package
• Distributed Mode GAT
• Parallel algorithm for concurrent execution
• Load balancing on multiple Servers

18
Performance Parallel Version
19
Performance Scalability
20
Summary

• GAT
• produces output consistent with existing data
  mining and analysis tools
• is faster than comparable tools
• scales to accommodate increasingly large files
• has a modular, extensible design
• has a distributed framework that can be applied
  to other methods of gene expression analysis
• is an open-source project

21
Future Work

• Incorporate a GUI
• Investigate scalability issues (MCBI)
• Add additional functions to the C statistical
  core

http//www.csis.gvsu.edu/den