1. Introduction to Microarray Gene Expression
Shyamal D. Peddada Biostatistics Branch
National Inst. Environmental
Health Sciences (NIH) Research Triangle Park, NC

2. Outline of the four talks
A general overview of microarray data
Some important terminology and background
Various platforms
Sources of variation
Normalization of data
Analysis of gene expression data - Nominal explanatory variables
Two types of explanatory variables
Scientific questions of interest
A brief discussion on false discovery rate (FDR) analysis
Some existing methods of analysis.

3. Outline of the four talks
Analysis of ordered gene expression data
Common experimental designs
Some existing statistical methods
An example
Demonstration of ORIOGEN
Some open research problems
Analysis of data from cell-cycle experiments
Some background on cell-cycle experiments
Modeling the data
Data from multiple experiments
Some open research problem

4. Talk 1: An overview of microarray data

5. To perform statistical analysis of any given data
It is important to understand all sources of (i) bias, (ii) variability.
Some basic understanding of the underlying technology!
Understand the sampling/experimental design

6. Some Important Terminology and Background

7. Central Dogma of Molecular Biology

8. Some background terminology: DNA and RNA
DNA (Deoxyribonucleic acid) - Contains genetic code or instructions for the development and function living organisms. It is double stranded.
Four Nucleotides (building blocks of DNA)
Adenine (A), Guanine (G),
Thymine (T), Cytosine (C)
Base pairs: (A, T) (G, C)
E.g ’ ---AAATGCAT---3’
3’ ---TTTACGTA---5’

9. Some background terminology: DNA and RNA
RNA (Ribonucleic acid) - transcribed (or copied) from DNA. It is single stranded. (Complimentary copy of one of the strands of DNA)
RNA polymerase - An enzyme that helps in the transcription of DNA to form RNA.
Four Nucleotides (building blocks of DNA)
Adenine (A), Guanine (G),
Uracil (U), Cytosine (C)
Base pairs: (A, U) (G, C)

10. Some background terminology: Types of RNA
Types of RNA - (transfer) tRNA,
(ribosomal) rRNA, etc.
mRNA - messenger RNA. Carries information from DNA to ribosomes where protein synthesis takes place (less stable than DNA).

11. Some background terminology: Oligos
Oligonucleotide - a short segment of DNA consisting of a few base pairs. In short it is commonly called “Oligo”.
“mer” - unit of measurement for an Oligo. It is the number of base pairs. So 30 base pair Oligo would be 30-mer long.

12. Some background terminology: Probes
cDNA - complimentary DNA. DNA sequence that is complimentary to the given mRNA.
Obtained using an enzyme called reverse transcriptase.
Probes - a short segment of DNA (about 100-mer or longer) used to detect DNA or RNA that compliments the sequence present in the probe.

13. Some background terminology: “Blots” - Origins of Microarrays
Southern blot (Edwin Southern, 1975 J. Molec. Biol.)
A method used to identify the presence of a DNA sequence in a sample of DNA.
Western blot (immunoblot)
to identify a specific protein from a tissue extract.

14. Some background terminology
Southwestern blot
to identify and characterize DNA-binding proteins.
Northern blot
A method used to study the gene expression from a sample of mRNA.

15. Microarrays …

16. Northern blot Vs Microarray
Rate of expression analysis
Thousands of genes at a time
(High throughput)
Few genes at a time
Automation
Automation possible
Manual
Scope
Allows to explore relationships among several 100’s of genes at the same time
Limited

17. What is a Microarray?
Sequences from thousands of different genes are immobilized, or attached, at fixed locations.
Spotted, or actually synthesized directly onto the support.

18. Microarray Technology
Two color dye array (Spotted array)
Spotted cDNA microarrays
Spotted oligo microarrays
Single dye array
In situ oligo microarrays

19. Microarray Technology

20. Spotted Microarrays

21. Spotted DNA Microarray
Slides carrying spots of target DNA are hybridized to fluorescently labeled cDNA from experimental and control cells and the arrays are imaged at two or more wavelengths
Expression profiling involves the hybridization of fluorescently labeled cDNA, prepared from cellular mRNA, to microarrays carrying thousands of unique sequences.

22. Spotted DNA Microarray
Spotted DNA array is typically “home made” so you need to think about:
cDNA or Oligo
Location of the Oligo in a given gene
Oligo length - number of bp?

23. Spotted DNA Microarray
Gene expression:
Y < 0; gene is over expressed in green labeled sample compared to red-labeled sample
Y = 0; gene is equally expressed in both samples
Y > 0; gene is over expressed in red-labeled sample compared to green labeled sample

24. Single Dye Microarrays

25. Major Commercial Platforms
More than 50 companies are currently offering various DNA microarray platforms, reagents and software
Affymetrix dominated the marker for many years
*Agilent has one and two-color microarray platform

26. Affymetrix GeneChip
Each gene is represented by 11 to 20 oligos of 25-mers
Probe: An oligo of 25-mer
Probe Pair: a PM and MM pair
Perfect match (PM): A 25-mer complementary to a reference sequence of interest (part of the gene)
Mismatch (MM): same as PM with a single base change for the middle (13th) base (G <-> C, A <-> T)
Probe set: a collection of probe-pairs (11 to 20) related to a fraction of gene

27. Affymetrix call for the presence of a signal
Affymetrix detection algorithm uses probe pair intensities to obtain detection p-value
Using this p-value they decide whether the signal is
“ present”, “marginal” or “absent”

28. Affy call Detection of p-value
Calculate Kendall’s tau T for each probe pair
T = (PM-MM) / (PM+MM)
Determine the statistical significance of the gene by computing the p-value.

29. Affy call
Ref: Affymetrix Technical Manual

30. Affymetrix Vs Illumina
Ref: Pan Du & Simon Lin

32. Which Platform to Choose?
Every platform has its unique feature
Choose platform based on
Nature of the study
Amount of available RNA
Cost
Platform comparison in MAQC study

33. MAQC Project
Objective: To generate a set of quality control tools for microarray research community
137 participants representing 51 organizations
Gene expression from two distinct RNA samples (total 4 samples)
Sample A = Universal Human Reference RNA(UHRR)–100%
Sample B = Human Brain Reference RNA(HBRR) – 100%
Sample C = 75% UHRR + 25% HBRR
Sample D = 25% UHRR + 75% HBRR

34. Microarray Data Analysis

35. Why Normalize Data?
To “calibrate”/adjust data so as to reduce or eliminate the effects arising from variation in technology and other sources rather than due to true biological differences between test groups.

36. Sources of bias/variation
Tissue or cell lines
mRNA
It can degrade over time - so there is a potential batch effect if portions of experiment are performed at different times
Purity and quantity
Dye color effect (spotted arrays)
Variation due to technology - is substantially reduced with improved technology
Etc.

37. A useful graphical representation of data
Data matrix:
Let

38. A useful graphical representation of data
Let its spectral decomposition be given by
where

39. A useful graphical representation of data
Then
Plot

40. Common Normalization Methods
Internal Control Normalization
Global Normalization
Linear Normalization (Spotted arrays)
Non-linear Normalization Method (Spotted arrays) - LOWESS curve.
ANOVA
COMBAT (for batch effect)

41. Internal control normalization (Housekeeping gene(s))
Expression of each gene is measured relative to the average of house keeping genes.
Basic assumption: Expression of housekeeping genes does not change.
Disadvantage:
House keeping genes may be highly expressed sometimes. Unexpected regulation of house keeping gene(s) leads to misinterpretation

42. Global Normalization Basic assumption Regression of
Mean/Median expression ratio of all monitored mRNAs is constant across a chip.
Regression of
In simple terms the log ratios are corrected by a common “mean” or “median”
This method can also be applied to single Dye data

43. Linear Normalization (for spotted arrays)
Basic assumption
Mean/Median expression ratio of all monitored mRNAs depends upon the average intensity
Regression of

44. Non-Linear Normalization (for spotted arrays)
Basic assumption
Mean/Median expression ratio of all monitored mRNAs depends upon the average intensity
Regression of
Where is estimated by the robust scatter plot
smoother LOWESS (Locally WEighted Scatterplot Smoothing)

45. Analysis of Variance (ANOVA)
Standard Analysis of Variance model
Response variable - Gene expression
Explanatory variables:
Dye color
Batch
Other potential effects?
Advantage: Statistically significant
genes can be identified while controlling for the
various experimental conditions/factors.

46. Some important experimental designs
Pooled Samples versus Separate samples
Sometimes there may not be sufficient biological sample/specimen from a given animal. In such cases biological samples are pooled from several identical animals to form a sample.

47. An example of a pooling design (for each treatment group)
Subjects Pool Observations
(Microarray chips)

48. The pooling design Subjects Pool Observations (Microarray chips) 9 3 6
(3 per pool)
More generally:
n p m
(r=n/p per pool)

49. The standard design Subjects # Pool Observations (Microarray chips)
(r=1)
More generally:
n p=n m=n

50. Some issues What are the underlying parameters?
Effect of pooling on power.
The basic assumption. Validity of the assumption.

51. Parameters
Total variation in the expression of a gene can be decomposed in to:
Biological variation
Technical variation
Biological samples (n)
Number of pools (p)
Biological samples per pool (r=n/p)
Observed number of samples (e.g. microarrays) (m)

52. Some comments about pooling
Variance of the estimated mean expression of a gene depends on:
number of pools (p)
number of bio samples per pool (r)
number of arrays (m)
biological variation
Technical variation.
Pooling works well when the biological variation in the gene
expression is substantially larger than the technical variation.

53. Power comparisons # Bio #Micro Pool size Power
5/group 5/group 1 (Standard design)
6/group 6/group 1 (Standard design)
6/group 3/group 2 (i.e 3 pools/group)
8/group 4/group (i.e. 4 pools/group)
10/group 5/group (i.e. 5 pools/group)
Zhang and Gant (2005)

54. Power comparisons Conditions of the simulation study:
Biological variation is 4 times the technical variation.
False positive rate is
Detect 2-fold expression.
Data are normally distributed.

55. A fundamental assumption
Biological averaging:
Suppose an experiment consists of pooling “r” samples. Then
the expression of a gene in the pooled sample is assumed to
be the average of the gene’s expression in the “r” samples.
This assumption need not be true especially if the expression
values are transformed non-linearly.

56. Some important experimental designs
Reference designs (Spotted array)
Each treatment sample is hybridized against a common reference control.
Loop designs (Spotted array)
Suppose we have a control and three experimental groups A, B and C. Then hybridize Control and A, A with B, B with C and C with A.

57. Data Analysis - Preliminaries
Normalization
Transformation of data (usual methods)
Perhaps first fit ANOVA and plot the residuals
Log transformation
Square root
More generally, Box-Cox family of transformations
Identify potential outliers in the data (again, perhaps use the residuals)

58. Data Analysis
Method of Analysis depends upon the scientific question of interest.
In the next three lectures we describe several general methods and illustrate some using real data!