1. Microarray Experiment Design and Data Interpretation
Susan Hester, Ph.D.
Environmental Carcinogenesis Division
Toxicogenomic Core Facility
US EPA

2. Presentation Outline Traditional biology versus genomics
Basics of genomics
Data mining goals and approaches using parallel analyses
-some examples
Interpreting changes in gene expression to identify altered molecular pathways
Evaluating pathway alterations in concert with traditional toxicology data for greater understanding of mode of action

3. Traditional Biology Measure one tree at a time Measure one element
in samples

4. “Omic” Biology Measure tens of thousands of elements in 2 to 4 samples
Measure Forests
(groups of trees)

5. Genomic research is a data-rich technology
Microarrays are called chips or arrays
Takes advantage of the natural property of DNA to pair with its complimentary strand
One strand is built into the array and then is used as a probe for the complementary strand in the biologic sample
The binding confirms the presence of mRNA or cDNA
In the sample

6. Genomic Profiling-Find ”Significantly Changed Genes”
From:
All probesets
Typical experiment
is ~ 1M datapoints
To:
Reduce to a much
smaller number
of “meaningful genes”

7. Finding genes in samples-1st step
1 genechip cell location
1 genechip
apply sample

8. 2nd step Tagged DNA fragments that base pair will glow 2nd step
shine light
final image
text file with
gene intensities

9. Experimental Design Use adequate controls Sample collection
Choose time-points and doses
Hybridization schemes-1 or 2 colors

10. Data Quality and Data Mining
RNA quality
Scans
Summary statistics

11. RNA quality: Agilent 2100 Bioanalyzer Measure RNA quality and quantity
Uses small sample size and take minutes
Good
Quality
RNA
Degraded
RNA
Agilent Gel Image

12. QC Assessment of Scanned Slide
Showing Good
Dynamic Range
of Signal Intensity
Low background
signal
Poor scan
Good scan

13. Summary Statistics for each array
Raw gene intensity distribution
for each array
After normalization shows
reduced variance
max
median
min
Grp

14. Example of with-in group outliers
Example of 2 array outliers
(high and low median values)
Arrays

15. Goals of Data Mining genes”
Reduce the large dataset by first exclude “unchanging
genes”
Early microarray papers used a simple “fold change”
to find differences
Most analyses now rely on statistical tests to identify
changed genes-supervised versus unsupervised
Find genes that distinguish the various biologic classes
“significant genes”

16. General Approach: From many genes to a few
28,000 rat genes
34,000 mouse genes
normalize data to compare across arrays
analysis begins here
supervised (prior knowledge) and unsupervised (no prior knowledge)
T test, ANOVA, etc PCA, KNN, clustering
genes…now associate with gene name
using databases to assign gene function
characterize genes into pathways
explore pathways by combining into networks

17. Array Image Inspection Confirms the Induction of Many Genes
1 uM As uM As

18. Statistical Filter shows more significant genes at higher doses
1 uM As
50 uM As
genes that have values>1.5 fold
and significant
p<0.05

19. Many Views of the Data Principal Component Analysis (PCA)
Table of filtered genes
Principal Component Analysis (PCA)
Venn Diagrams-gene level
Correlate Transcription with Functional Assays
Map genes to pathways
Venn Diagram-pathway level

20. Table view: Significantly Altered Genes by Chemical, Day and Dose
in rat liver

21. Principal Component Analysis
Identifies dose-response, if present
Assess experiment
Worth analyzing ?
Identify outliers-bad chips
Find samples with similar expression patterns
What it does
What it looks like:
uses all samples and genes
using statistics, reduces and plots the data
helps visualize data in 2 or 3 planes (3D)
What it tells
groups samples or genes with similar profiles
differentiates treatment or exposure groups

22. Principal Component Analysis Rat Liver

23. Numbers of Common and Unique Genes Over Time (High Dose)-rat liver

24. Dose response corresponds to functional assays
Better description of dose response by genomics

25. Mapping genes to pathways
Process
p-Value
# of genes Expressed
# of genes in Pathway %
Transcription of Retinoid-Target genes
Cell signaling/Regulation of transcription
7.56E-09 68
125 54
Regulation activity of EIF2
Cell signaling/Translation regulation
5.86E-05 31 56 55
IGF-R signaling
Growth and differentiation
4.57E-06 40 72
AKT signaling
9.50E-06 33 57 58
PTEN pathway
3.65E-05
Tryptophan metabolism
Metabolic maps/Amino acid metabolism
3.99E-05 17 24 71
Cholesterol Biosynthesis
Metabolic maps/Steroid metabolism
6.25E-06 16 22 82
GTP-XTP metabolism
Metabolic maps/Nucleotide metabolism
4.58E-07 34 63
CTP/UTP metabolism
1.32E-05 60
ATP/ITP metabolism
1.49E-05 36 65

26. Pathway Venn
Unique and common pathways over time

27. Pathway and network visualizations
cellular
molecular
network
metabolic
transcription

28. Example of a molecular pathway with gene intensity values added
Oxidative Phosphorylation pathway
red=gene induced
green=gene repressed
rainbow=mixed
ATPase
Oxidoreductase
NADH dehydrogenase
succinate dehydrogenase
complex
cytochrome c
oxidase subunit

29. Cellular pathway extracellular cytoplasmic nuclear Note c-Jun
JNK1, ERK1
repression*
nuclear
Expression legend
Green= decreased
Red=increased
Rainbow=mixed

30. Gene Network:
One Transcription factor:

31. Network objects mapped to cellular localization

32. Conclusions Steps for a successful microarray experiment:
Experiment design-focus your research question
Data quality assessment
Supervised and unsupervised analyses
Integrating gene expression results with other
phenotypic endpoints