Transcript Document

Microarrays
A snapshot that captures the activity
pattern of thousands of genes at once.
Custom spotted arrays
Affymetrix GeneChip
Practical Applications of Microarrays
Gene Target Discovery
By allowing scientists to compare diseased cells with normal cells, arrays can
be used to discover sets of genes that play key roles in diseases. Genes that
are either overexpressed or underexpressed in the diseased cells often
present excellent targets for therapeutic drugs.
Pharmacology and Toxicology
Arrays can provide a highly sensitive indicator of a drug’s activity
(pharmacology) and toxicity (toxicology) in cell culture or test animals.
This information can then be used to screen or optimize drug candidates
prior to launching costly clinical trials.
Diagnostics
Array technology can be used to diagnose clinical conditions by detecting
gene expression patterns associated with disease states in either biopsy
samples or peripheral blood cells.
Microarray Platforms
•Oligonucleotide-based arrays
•25mers spotted on a glass wafer, Affymetrix
GeneChip arrays
•Custom spotted 50-80mers generated from
known sequences.
•cDNA
•Inserts from cDNA libraries
•PCR products generated from gene specific or
universal primers
GeneChip Instrument System
®
Fluidics
Station
Scanner made by
Hewlett-Packard
Computer
Workstation
GeneChip Probe Array
®
GeneChip Probe Arrays
®
GeneChip Probe Array
Hybridized Probe Cell
Single stranded, fluorescently
labeled DNA target
Oligonucleotide probe
1.28cm
*
*
*
*
*
24µm
Each probe cell or feature contains
millions of copies of a specific
oligonucleotide probe
Over 250,000 different probes
complementary to genetic
information of interest
Image of Hybridized Probe Array
Synthesis of Ordered
Oligonucleotide Arrays
Light
(deprotection)
Mask
OOOOO
TTOOO
HO HO O O O
T–
Substrate
Light
(deprotection)
Mask
CATAT
AGCTG
TTCCG
TTCCO
TTOOO
C–
Substrate
REPEAT
Probe Tiling Strategy
Gene Expression
(25-mer)
Gene Expression
Tiling Strategy
[
]
[
]
[
[
Uninduced
]
]
[
]
[
]
Induced
40 separate hybridization events are involved in
determining the presence or absence of a transcript
80 separate hybridization events are involved determining
differential gene expression of a transcript between two
samples
Starting material for Microarrays
Platform
Affymetrix
Poly (A)+ mRNA
Total RNA
~2 mg
~10 mg
Spotted arrays
Poly (A) +
Total RNA
~0.4 – 2 mg
10 -100 mg
Agilent Bioanalyzer 2100
Total RNA
Fragmented cRNA
Experimental Design
Biotin - labeled
cRNA transcript
Cells
B
+
Poly (A)
RNA
Or
Total RNA
IVT
cDNA
Biotin-UTP
Biotin-CTP
B
B
B
B
B
B
Fragment
heat, Mg2+
B
Hybridize
B
B
B
B
Wash & Stain
Scan
(8 minutes)
(75 minutes)
(16 hours)
Biotin - labeled cRNA
fragments
Add Oligo B2 &
Staggered Spike
Controls
.DAT file and .CEL file
.CHP file
Absolute Analysis
Normalization and Scaling
Non-biological factors can contribute to the variability of data
in many biological assays, therefore it is important to minimize
the non-biological differences. Factors that may contribute to
variation include:
•Amount and quality of target hybridized to array
•Amount of stain applied
•Experimental variables
The data can be normalized from:
•a limited group of probe sets
•all probe sets
Thus the normalization of the array is multiplied by a
Normalization Factor (NF) to make its Average Intensity
equivalent to the Average Intensity of the baseline array.
Normalization and Scaling
Average intensity of an array is calculated by averaging all
the Average Difference values of every probe set on the array,
excluding the highest 2% and lowest 2% of the values.
Background Calculation: Measure of non-specific
fluorescence attributed to hybridization conditions and
sample.
Defaults - 16 sectors
Horizontal (HZ) : 4
Vertical (VZ) : 4
Background
• Probe Cells with the lowest 2 %
intensity values for each sector
are averaged.
Probe
Array
•
This value is subtracted from all
cell intensities in each sector
before further analysis.
Signal Noise
Noise results from small variations in the digitized signal observed
by the scanner as it samples the probe array’s surface. The level of
the noise is calculated by the software, and then used as one of the
criteria to determine the significance of differences between PM
and MM probe cells, and differences in probe set intensities across
two probe arrays.
Noise Calculation:
Q
Pixel to pixel variation determined from background
Q  1/ N

stdev i
iallbgcells
Total # of background cells
- lowest 2% for each sector.
Noise for each sector
of a given probe array
pixel i
 SF  NF
Total # of pixels
in a feature i
Standard deviation of
the intensities of the
pixels making up
feature i
Normalization
Factor
Scaling
Factor
What determines a positive
or negative probe pair?
Positive Probe Pair
Negative Probe Pair
1) PM - MM > SDT
1) MM - PM > SDT
2) PM / MM > SRT
2) MM / PM > SRT
and
PM is more than MM.
Yes, this probe pair
is detecting a signal.
and
PM and MM are similar. More MM than PM.
No differential signal
Signal is not specific to
detected.
targeted sequence.
Statistical Difference Threshold
PM - MM > SDT
• Calculated by the software based on the
noise (Q):
SDT = (Q) x (SDTmult)
SDTmult (multiplier) is set by default to 2.0 when the single
SAPE staining protocol is used (usually with 50m feature arrays), and
4.0 when the antibody amplification protocol is used (usually with
24m feature arrays).
(SDTmult) can be modified by user:
- increasing makes the analysis more stringent;
decreasing less stringent
Statistical Ratio Threshold
PM / MM > SRT
• SRT is set by user
– increasing makes the analysis more stringent;
decreasing less stringent
• Default SRT value is 1.5
– an SRT threshold value of 1.5 means that the
intensity of the PM must be 50% greater than
MM (after background subtraction) to meet
criteria
Probe Pairs in Average
Used in calculation of Log Average Ratio and
Average Difference
Pairs in Average
A “Trimmed” probe set prevents outlier probe pairs (extremely
positive or negative) from inclusion in calculations for Log
Average Ratio (and Average Difference)
8 probe pairs or fewer:
Greater than 8 probe pairs:
• All probe pairs are
• Super Scoring takes
used
place
Super Scoring
Used in calculation of Log Average Ratio and Average Difference

A mean and a standard deviation are calculated for the intensity
differences among an entire probe set.

A filter is then applied to each member of the probe set.

Probe pairs outside of the number of standard deviations set in the
parameters are excluded from the calculations of Log Average Ratio
and Average Difference
STP is the parameter for setting the number of standard deviations
used in Super Scoring. Default Setting is 3 (excludes everything
outside of 99.7% of the mean)
Positive Fraction
Number of positive probe pairs/total pairs used
15/20 = 0.75
Log Average Ratio
Log Avg = 10 * [ log (PM /MM)]/(# Probe Pairs in Average)
An average of the log of the intensity ratios is
calculated for each probe set from the pairs in
average and multiplied by 10.
Positive/Negative Ratio
Ratio of positive probe pairs to Negative probe
pairs in a probe set
Pos/Neg = 18/2 = 9
Average Difference
Avg Diff =  (PM - MM) / Pairs in Average
Average difference is calculated by taking the difference between
PM and MM of every probe pair and averaging the differences over
the entire probe set.
Average difference correlates with expression level
Average Difference is not used in the
Absolute Call Decision Matrix
Absolute Call
Decision Matrix - Absolute Analysis Threshold Values
Present
Marginal
4.0
0.43
1.3
Max
3.0
0.33
0.9
Min
Absent
Pos/Neg Ratio
Positive
Fraction
Log Avg Ratio
Calls must be in the Present bin in order for
quantification metrics to be informative
Comparison Analysis
Increased Probe Pairs & Decreased
Probe Pairs
Increased
Probe Pair
Neither
Increased
or
Decreased
Decreased
Probe Pair
Increased Probe Pair
(PM - MM)exp - (PM - MM)base > Change threshold (CT)
And
[(PM - MM)exp - (PM - MM)base] / (PM - MM)base > (PCT)/100
Probe Set on Baseline Probe Array
Probe Set on Experimental Probe Array
Increased Probe Pairs
Compares changes in relative intensity between two probe
pairs on two probe arrays, not positive/negative probe pair
changes
Decreased Probe Pair
(PM - MM)base - (PM - MM)exp > Change threshold (CT)
And
[(PM - MM)base - (PM - MM)exp] / (PM - MM)base > (PCT)/100
Probe Set on Baseline Array
Probe Set on Experimental Array
Decreased Probe Pair
Compares changes in relative intensity between two probe pairs
on two probe arrays, not positive/negative probe pair changes
Thresholds used in comparison analysis
Change Threshold (CT)
• The CT can be calculated in either of two ways:
• Calculated by the software, based on the SDTs of
the two probe arrays being compared
• Calculated as the product of a parameter called
CT multiplier (CTmult) and Q.
 SDT 2  SDT 2 - appliedif input field is blank
exp
bl


CT  
OR

CTmult  maxQexp , Qbl  - appliedby settingCTmult

CTmult is a default setting (80) or can be set by the user
Percent Change Threshold (PCT)
User Defined (default 80); means a probe pair must change
80%
Increase or Decrease Ratio
Increase Ratio = # Increased probe pairs / # probe pairs used
Decrease Ratio = # Decreased probe pairs / # probe pairs used
Probe Set on Baseline Array
Probe Set on Experimental Array
10 Increased Probe Pairs / 20 = 0.5
Compares changes in relative intensity between two probe pairs on
two probe arrays, not positive/negative probe pair changes
Max (Increase/PP used),(Decrease/PP used)
Calculates the number of probe pairs that have changed
in a certain direction.
Increase/PP used = number of increased probe pairs/number of
probe pairs used
Decrease/PP used = number of decreased probe pairs/number of
probe pairs used
Max Inc & Dec = Max (0.95, 0.05) = 0.95
This larger of the values will be used in the decision matrix, which
determines whether each transcript’s expression level has changed
between baseline and experimental.
Increase/Decrease Ratio
Ratio of increase probe pairs over decreased probe pairs
Increased: 6
Decreased:1
Inc/Dec = 6/1 = 6
Dpos - Dneg Ratio
Dpos-Dneg Ratio = positive change - negative change / # pp used
Positive Change =
#
Positive Probe Pairsexp -
#
Negative Change = # Negative Probe Pairsexp -
Positive Probe Pairsbase
#
Negative Probe Pairsbas
• Dpos - Dneg Ratio flags and excludes probe sets that change in
two directions () within one transcript. It also accounts for
changes in the neither bin.
Probe Set on Baseline Array
7 Positive PP
Example:
3 Negative PP 10 Neither
Probe Set on Experimental Array
14 Positive PP 4 Negative PP
Positive Change = (14 - 7) = 7
Negative Change = (4 - 3) = 1
2 Neither
Dpos -Dneg Ratio = (7 - 1) / 20 = 0.3
Log Average Ratio Change
• Log Average is recomputed for each probe set based
on probe pairs used in both the baseline and
experimental probe arrays.
Log Avg Ratio Change = Log Avgexp - Log Avgbase
Example:
Probe Set on Baseline Array
Probe Set on Experimental Array
2 probe pairs Not in
1 probe pair Not in
Average
Average
Total = 3 probe pairs Not in Average
Log average is recomputed for each probe set to take into
account any probe pairs that have been dropped (not in
average) or masked
Differential Call
Comparison
Analysis Threshold Values
-
Increase
No Change
Decrease
4.0
0.33
0.2
0.9
Max
3.0
0.43
0.3
1.3
Min
Inc/Dec
Ratio
Inc/Total
Dpos/Total
 Log
Avg Ratio
Calls must be in the Increase or Decrease bin in order
for quantification metrics to be informative
Average Difference Change
• Average Difference is recomputed for each probe set based
on common probe pairs to take into account any probe pairs
that have been dropped (not in average) or masked
Avg Diff Change =  Avg Diffexp - Avg Diffbase
Average difference change correlates to changes in
expression level
Average Difference Change is used in Fold Change
calculations, but not used in the Comparison Call
Decision Matrix
B = A
When an “ * ” is present in this column, it signifies that
In the baseline array, this transcript was absent.
Example:
Absolute call
A
B = A
*
Diff Call
I
Define
not significant, slight increase from
A baseline to A experimental
A
*
D
not significant, slight decrease from
A baseline to A experimental
P
*
I
IMPORTANT, increase in gene
expression, A baseline to P exp.
Fold Change: Measure of the relative change
in mRNA expression levels between experiments.
FC =
Avg Diff Change (exp-base) (recomputed)
max[ min (Avg Diff
exp,Avg
Diff
base),
QM * Qc]
Lesser of the two values
if (QM x QC) of either array
is greater than the average
difference of the transcript in
either the baseline or
experimental, the fold change
is calculated using the Noise.
In this case the fold change
is preceded by a (~) and
considered an approximation.
Defined by the
library file
+1
or
-1
AvgDiffexp  AvgDiffbase
AvgDiffexp  AvgDiffbase
Greater of the scaled or
normalized Qexp or Qbase
Sort Score
Based on a calculation that basically multiplies Fold
Change and Average Difference Change. The larger the
Sort score, the more further away the values are from
the noise.
Example:
Avg. Diff (baseline): 10
Avg. Diff (experimental) :100
Avg. Diff Change : 90
Fold change:10
Avg. Diff (baseline):100
Avg. Diff (experimental): 1000
Avg. Diff Change :900
Fold Change: 10
The fold change in both experiments is 10; however the Sort Score will be
approximately 10 times larger in experiment #2 than #1, due to
higher average difference change.
A fold change with a high sort score means that the average difference change
is relatively large.
Data Analysis
Save as *.txt file and import into
other statistical software programs
Data Visualization:
Data visualization is an important technique for gaining a
fundamental understanding of results of a microarray experiment.
you can detect outliers/anomalies, overall trends, clusters,
correlations with the following visual techniques.
• 1-D Profile Plots - e.g. time series response data
• Histogram / Frequency Plots - analyze distribution of gene expression data
• Star Plots - signature analysis of gene expression profiles
• Intensity Plots – color genes by gene expression across all exp’ts at once
• Scatter Plots – allow you to visualize high-dimensional microarray data
Integration:
Integrated into your environment by reading files in standard
formats and writing the results out in standard formats.
• Import flat files, comma or tab separated Formats, or URL’s
•Import from ODBC Data Source
•Tiles Saved in Portable Comma Separated (.csv) Format
•Automate Via Embedded Tcl Scripting Language
•Link to Other Applications by Selecting Data in Spreadsheet or in Graphics.
Data Processing:
•Analytical Spreadsheet can Handle Millions of Rows or Columns
•Scaling & Normalization (e.g. standardize, log-scale, log & linear scale, power)
•Sort rows by Value or by Similarity to Prototype (find genes most similar to
specified prototype)
•Missing Data Handling (e.g. analysis, casewise deletion, imputation)
Exploratory Analysis:
New, unexpected discoveries are most easily made during the exploratory
analysis stage.
•Cluster analysis – identify genes with similar expression profile
•Principal Components Analysis – visually and numerically analyze the correlation
inherent in the data (similarity of genes, of experiments)
•Multidimensional Scaling – visually analyze similarity of genes, tissues, or time
points using any one of 20 measures of similarity.
•Linear, Non-linear correlation – find significantly correlated genes, tissues,
or time points.
•Parametric & Nonparametric tests (e.g. chi-square, t-test, anova,
kolmogorov-smirnov) – genes that are significantly different
•Correspondence Analysis – measure the correspondence between (for example)
a cluster analysis grouping and a known functional class of genes.
•Randomization Experiments & Permutation Tests – evaluate likelihood of
chance
More Analysis Features
Correlations
•Find Genes with expression profiles similar to a chosen Gene
•Find Expression Profiles similar to a drawn Profile
•Multiple Ways to Define 'Similar' in the 'Find Similar' Search
Quantitative Restrictions
•Filter data by degree of expression or x-fold comparisons across experiments.
Find Interesting Genes Function Pathways
•Identify Potential New Candidates
Modeling:
Modeling tools allow us to make use of discoveries to build predictions on new,
unknown data. For example, we can build a neural network (or other predictive
model) capable of accurately categorizing tissue type or condition, based on
just one, two, or a few genes alone.
•Variable Selection – find the few genes needed to represent a profile for a
particular phenotype.
•Neural Networks – build a neural network to categorize new samples based
on the gene expression of the few selected genes.
•Discriminant Analysis (linear, quadratic) - build a statistical model to
categorize new samples based on the gene expression of few selected genes.
•Automated Model Validation - use cross-validation, jackknife, or bootstrap to
estimate the accuracy of your sample identification model
•Save Models for Prediction of new observations – save your predictive model
for actual use.
Cluster and Tree View
Microarray Process
Indirect labeling
Simple, highly sensitive technique
requires less starting RNA, and
creates evenly labeled DNA
without dye bias.
•Uniform incorporation of
fluorescent dyes produces more
reliable signals
•High sensitivity to detect lowcopy signals
•Requires only 10 to 20 µg of
total RNA or 0.4 to 1 µg of polyA
RNA
Clontech
Atlas™ Glass Fluorescent Labeling Kit
Stratagene
FairPlay™ Microarray Labeling Kit
Products used for spotting
Easy-To-Spot™ Products (Incyte Genomics)
•Every clone is sequence-verified prior to PCR
• PCR products are purified to remove excess salts,
unincorporated nucleotides, primers, and particulates
• Quality controlled production process with failure rate1 of less
than 10%
• 8,734 PCR products from sequenced-verified clones from the
UniGene database from NCBI, average length is greater than
500 nucleotides
•Between 1-3 ug of DNA per well. Enough to fabricate 500 to 1,000
arrays
• Corresponding clones available for purchase for further research
Array Ready Oligo set
(Operon Technologies)
Complete Yeast Genome Oligo Set
• Optimized 70-mer oligonucleotides for each of the 6,307 open reading
frames (ORFs) of Saccharomyces cerevisiae from the Saccharomyces
Genome Database (SGD) at Stanford University
•The amount of sample provided with each set is sufficient to print
between 2000 and 6000 slides, depending on the printing procedure
used.
Human Genome Oligo Set
•This Array-Ready Oligo Set™ contains arrayable 70-mers
representing 13,971 well-characterized human genes from the
UniGene database. This database is located at the National Center
for Biotechnology Information.
•All 70-mer oligonucleotides in the Human Genome Oligo Set were
designed from the representative sequences in the UniGene
database, Hs build #119. The set also contains 29 controls.
GeneMachine Omni Grid Arrayer
Printing Pin
Axon GenePix4000A Scanner
• 10mm pixel size
• Simultaneously scans array
slides at two wavelengths
• User-selectable laser power
• User-selectable focus poisitions
GenePix Pro Features
• Auto Align
Before Auto Align
After Auto Align
GenePix Pro Features
•Feature Viewer
P = pixel intensity
F = feature intensity
B = background intensity
Rp = ratio of pixel intensities
Rm = ratio of means
mR = median of ratios
rR = regression ratio
GenePix Pro Features
•Feature Pixel Plot
GenePix Pro Features
•Histogram
GenePix Pro Features
•Scatter Plot
GenePix Pro Features
• Results
Spotted glass slide microarrays
Advantages
Low cost per array
Custom gene selection
Any species
Competitive hybridization
Open architecture
Disadvantages
Clone management
Clone cost
Quality control
Affymetrix GeneChip system
Advantages
Stream line production
Large number of genes and ESTs/chip
Several number of species
Disadvantages
System cost
GeneChip cost
Propietary system
Limits on customizing
http://genome-www4.stanford.edu/cgi-bin/SMD/source/sourceSearch
The Stanford Online Universal Resource for Clones and ESTs
(SOURCE) compiles information from several publicly
accessible databases, including UniGene , dbEST , SwissProt
, GeneMapp99, RHdb , GeneCards and LocusLink. The
mission of SOURCE is to provide a unique scientific resource
that pools publicly available data commonly sought after for
any clone, GenBank accession number, or gene. SOURCE was
specifically designed to facilitate analysis of the large data
sets that biologists can now produce using genome-scale
experimental approaches.
Choose organism:
Choose search option:
Enter a search term:
Use a wildcard character (*) at the end and/or beginning of
the term to broaden your search.Choose type of information
to display: GeneReport: Gene Information (limited to those
in UniGene)
CloneReport: cDNA Clone Information (limited to those in
dbEST)
GeneCards Database
http://207.123.190.10/
Dragon Database
• annotate microarray data sets with biological information
• search for genes or proteins that have shared characteristics
• compare the genes represented on multiple array technologies
Dragon View
• view annotated microarray data
Dragon Map
• explore the map of human gene expression
Challenges in analyzing Microarray Data
•Amount of DNA in spot is not consistant
•Spot contamination
•cDNA may not be proportional to that in the tissue
•Low hybridization quality
•Measurement errors
•Spliced variants
•Outliers
•Data are high-dimensional “multi-variant”
•Biological signal may be subtle, complex, non linear,
and buried in a cloud of noise
•Normalization
•Comparison across multiple arrays, time points, tissues,
treatments
•How do you reveal biological relationships among genes?
•How do you distinguish real effect from artifact?
Factors to consider in designing
microarray experiments
•Need to do lots of control experiments-validate method
•Do replicate spotting, replicate chips, and reverse labeling
for custom spotted chips
•Do pilot studies before doing “mega chip” experiments
•Don’t design experiment without replication; nothing will
be learned from a single failed experiment
•Design simple (one-two factor) experiments,
i.e. treatment vs. untreatment
•Understand measurement errors
•In designing Databases; they are useful ONLY if quality
of data is assured
•Involve statistical colleagues in the design stages of your studies
Once you have identified an interesting
expression pattern, what comes next?
•With some arrays it is possible to purchase clones of interest for
further experimentation.
•Confirm that the particular clone you now have in your hand shows
the expression pattern so indicated by the array, quantitating
individual mRNA species.
•RT-PCR, Relative, quantitative RT-PCR uses an internal
standard to monitor each reaction and allow comparisons
between different reactions to be made.
• Competitive RT-PCR --a competition between a known amount
of a template and an unknown target.
•Northern analysis