Microarray Data Analysis

1
Microarray Data Analysis - The Need for
Normalization Christine Steinhoff Max Planck
Institut für Molekulare Genetik Berlin, Germany
2
Types of Arrays
Red/Green experiments
Affymetrix Chips
Radioactive filters
3
Data Analysis - Procedure
?
4
Data Analysis - Procedure
5
Example Quality Check
Ratio of intensitities of both channels
Yang, YH et al, SPIE BiOS, San Jose 2001
Product intensitity of both channels
6
Example Quality Check
Ratio of intensitities of both channels
Yang, YH et al, SPIE BiOS, San Jose 2001
Product intensitity of both channels
7
Data Analysis - Procedure
Starting with Image Processing Scanneroutput
informations about spot intensities local
background pins PCR plates localization
standard deviation...
Quality Check Are there any effects due to
pins PCR plates local effects ...
8
What is Normalization ?
Systematic Variation in Microarray Experiments
- Saturation (Scanner Labelling) -
Nonlinearity of Cy5, Cy3 Labelling -
Efficiencies of Cy5, Cy3 Labelling - Variation
of Low-Intensities - Pins - PCR Plates - Local
Effects ...
Normalization is the process of describing and
removing such variation
9
What is Normalization ?
10
Why do we normalize ? Do we need normalization ?
Goal Reliable Measurement of Ratios Patient
vs. Control Patient(red)/Control(green)
Patient(green)/Control(red)
In Self-Self-Hybridization we would
expect green/red 1 for all genes
Mixture of Unequal Labelling Noise not constant
Variance Differential Expression (not in this
example!) ...
11
Normalization Methods
Entire Dataset Overall Mean, Median,
Shorth Overall Regression Local
Regression Zscore ANOVA Variance
Stabilization useful for Most Genes
Unchanged- Settings
User Defined Sets Housekeeping (?!) Controls
etc useful for Most Genes Changed- Settings
12
Normalization Strategies
Local Regression determine regression lines
locally
13
Normalization Strategies
14
Comparison of Normalization Strategies
1 maximal differential genes (red, 138 genes)
discarding 5 lowest expressed genes (green, 691
genes) before log product vs. log ratio of
normalized intensities
No Normalization
Linear Regression
Local Regression
Overall Median
Zscore
ANOVA
Variance Stabilization
15
Comparison of Normalization Strategies
Goal Detection of differentially expressed
genes Set of 30 maximal differential genes out
of 13824
16
Comparison of Normalization Strategies
17
Comparison of Normalization Strategies
Goal Detection of Differentially expressed Genes
Var Stab ANOVA Lin Regr Least Med Local
Regr Mean Median Shorth Zscore Raw
18
What is the right Normalization-method to use ?
General Assumption Most Genes Unchanged!
- Quantifiable Differential Expression ! - high
variance in low intensities - removing spatial
effects of various types (pins, PCR plates,
nonlinearity of labelling) - overall nonlinear
data - analyzing for a variety of influences -
specific for Dye-Swap-setting
Variance Stabilization Local
Regression ANOVA
19
Summary

• Quality Check can detect removable effects which
  otherwise could lead to false
• positives
• Normalization of the data is essential to
• (1) remove
  systematic effects and
• (2) make
  experiments comparable
• Different Normalization Methods can lead to
  different interpretations of the data
• Depending on the experimental question different
  methods are appropriate for
  quantifiable expression differences for ex.
  Variance Stabilization
• for removal of spatial effects and overall
  nonlinear data for ex. Local Regression
• for detection of various influences in Dye
  Swap settings for ex. ANOVA
• For Red/Green - experiments always use Dye
  Swaps

20
Acknowledgement
Ulrike Nuber H.-Hilger Ropers Martin Vingron
Human Molecular Genetics Computational
Molecular Biology