Download Analyzing Microarray Data with Methods from Statistics and

Analyzing Microarray Data
with Methods from
Statistics and Machine Learning
B-IT IPEC Winter School 2008
Prof. Dr. A. B. Cremers
Jörg Zimmermann
DNA Microarray Data
• Genome Chips containing a collection
of microscopic DNA spots
• Simultaneous determination of > 105
Gene Expression Levels
• Dramatic acceleration of data
aquisition
• New possibilities for disease diagnosis,
treatment studies, network analysis, …
Analyzing Microarray Data with Methods from Statistics and Machine Learning
DNA Microarray Data
The resulting data have the form:
x11 x12 … x1n ( L1 )
.
.
.
.
.
.
xp1 xp2 … xpn ( Lp )
n = number of measured cell states (e.g. gene expression levels)
p = number of samples
xij = real number e.g. representing expression level of gene j in sample i
Li = Label of sample i
Analyzing Microarray Data with Methods from Statistics and Machine Learning
Challenges for Data Analysis
• Normalization (removing systematic measurement effects)
• Variable Selection (Identification of relevant Variables)
• Large sample Effects:
Type I and Type II errors (False positives / False negatives)
• Dimensionality Reduction
• Identification of new disease classes
• Classification of data into known disease classes
Analyzing Microarray Data with Methods from Statistics and Machine Learning
Cluster Analysis
Finding Structure in data without labels (unsupervised learning)
Does a cluster characterize a (new) disease type?
Analyzing Microarray Data with Methods from Statistics and Machine Learning
Prediction Problem
• Classify data into known disease classes:
Supervised Learning
• Split
data in Training and Test set
• Learn a model on the training set
• Evaluate model on the test set
Analyzing Microarray Data with Methods from Statistics and Machine Learning
Prediction Problem
Under- and Overlearning:
Analyzing Microarray Data with Methods from Statistics and Machine Learning
Data Analysis Methods
Dimension Reduction
• PCA (Principle Component Analysis)
• ICA (Independent Component Analysis)
• Multidimensional Scaling
Unsupervised Learning
• K-Means / K-Medoid
• Hierarchical Clustering Algorithms
Supervised Learning
• Linear Discriminant Analysis
• Maximum Likelihood Discrimination
• Nearest Neighbor Methods
• Decision Trees
• Random Forests
Analyzing Microarray Data with Methods from Statistics and Machine Learning
Organisation
Schedule:
31.3.2008 – 4.4.2008, B-IT Building
Language: german and english (Slides in english)
Talk: 45 min + 15 min discussion
Documentation: 10 – 15 pages (german or english)
Bereich (DPO Bonn): B
Background Literature:
Hastie, Tibshirani, Friedman:
The Elements of Statistical Learning, Springer, 2001
Contact: [email protected]
Summer Course: Gene Mining and Network Analysis
Summer School: Programming Data Analysis Algorithms with R
Analyzing Microarray Data with Methods from Statistics and Machine Learning