Effective Feature Selection Framework for Cluster Analysis of Microarray Data

1
Effective Feature Selection Framework for
Cluster Analysis of Microarray Data

• Gouchol Pok
• Computer Science Dept.
• Yanbian University
• China

Keun Ho Ryu DB/Bioinformatics Lab Chungbuk Natl
University Korea
2
Outline

• Background
• Motivation
• Proposed Method
• Experiments
• Conclusion

3
Feature Selection

• Definition
• Process of selecting a subset of relevant
  features for building robust learning models
• Objectives
• Alleviating the effect of the curse of
  dimensionality
• Enhancing generalization capability
• Speeding up learning process
• Improving model interpretability

from Wikipedia http//en.wikipedia.org/wiki/Featu
re_selection
4
Issues in Feature Selection

• How to compute the degree to which a feature is
  relevant with the class (discrimination)
• How to decide if a selected feature is redundant
  with other features (strongly correlated)
• How to select features so that classifying power
  is not diminished (increased)

• Removal of irrelevancy
• Removal of redundancy
• Maintain class-discriminating power

5
Selection Modes

• Univariate method
• considers one feature at a time based on score
  rank
• measures are Correlation, Information measure,
  K-S statistic, etc
• Multivariate method
• considers subsets of features altogether
• Bayesian and PCA based selection
• in principle, more powerful than univariate
  method, but not always in practice (Guyon2008)

6
Hard Case in Univariate method (Guyon2008)
Adopted from Guyons tutorial at IPAM summer
school
7
Proposed method Motivation

• Method that fits 2-D microarray data
• typical forms thousands of genes (rows) and
  hundreds of samples (columns)
• Multivariate approach
• Feature relevancy and redundancy are addressed
  simultaneously

8
System Flow
samples
genes
9
System Flow (cont.)
10
Methods Step1

• Perform column-based difference op.
• Di(N,M) C(N,M) ? Ci(N,1), i 1,2,, M
• Difference operator may depend on applications,
  e.g. Euclidean or Manhattan distance
• Di(N,M) contains class-specific info. w.r.t each
  gene

genes
11
Methods Step2

• Apply thresholds
• Find kind of emerging patterns which contrast 2
  classes
• Suppose 1, 2,, j ? C1 and j1, j2, M ? C2
• Sort the values in each column of Di(N,M)
• 25-threshold to the same class differences and
  75-threshold to the different class differences

C1
C2
C1
C2
C1
C2
25
75
12
Methods Step3

• Extract class-specific features
• Within-class summation of binary values (count
  1s)

13
Methods Step4

• Gene selection
• Apply different threshold value for different
  class
• Gene selection we are done for the row-wise
  reduction

threshold
14
Methods Step5

• Column-wise reduction by clustering
• Classification of samples
• Applied NMF method

15
Nonnegative Matrix Factorization (NMF)

• Matrix factorization A VH
• A n ? m matrix of n genes and m samples.
• V (n ? k) k columns of V are called basis
  vectors
• H (k ? m) describes how strongly each building
  block is present in measurement
  vectors

16
NMF Parts-based Clustering (Brunet2004)

• Brunet introduce meta-genes concept

17
Experiments Datasets

• Leukemia Data
• 5000 genes
• 38 samples of two classes
• 19 samples of ALL-B and 8 samples of ALL-T type,
• 11 samples of AML type.
• Medulloblastoma Data
• 5893 genes
• 34 samples of two classes
• 25 classic type and 9 desmoplastic
  medulloblastoma type
• Central Nervous System Tumors Data
• 7129 samples
• 34 samples of four classes
• 10 classic medulloblastomas, 10 malig-nant
  gliomas, 10 rhabdoids, and 4 normals

18
Classification

• Given a target sample, its class is predicted by
  the highest value in k-dim column vector of H

19
Results

• Leukemia Data (ALL-T vs. ALL-B vs. AML)

20
Results

• Medulloblastoma Data (Classic vs. Desmoplastic)

21
Results

• Central Nervous System Tumors Data (4 classes)

22
Conclusions Future work

• Our approach tries to capture a group of
  features, but in contrast to holistic methods
  such as PCA and ICA, intrinsic structure of data
  distribution is preserved in the reduced space.
• Still, PCA and ICA can be used as an aide to look
  into the data distribution structure, and provide
  useful information for further processing to
  other methods.
• Our on-going research is on how to combine the
  PCA and ICA to the proposed work

23
References

• Wikipedia, http//en.wikipedia.org/wiki/Feature_se
  lection
• J.-P. Brunet, P. Tamayo, T. Golub, and J. P.
  Mesirov. Metagenes and molecular pattern
  discovery using matrix factorization. PNAS,
  101(12)4164-4169, 2004.
• L. Yu and H. Liu. Feature selection for
  high-dimensional data A fast correlation-based
  filter solution. In Proc 12th Int Conf on Machine
  Learning (ICML-03), pages 856863, 2003
• Biesiada J, Duch W (2005), Feature Selection for
  High-Dimensional Data A Kolmogorov-Smirnov
  Correlation-Based Filter Solution. (CORES'05)
  Advances in Soft Computing, Springer Verlag, pp.
  95-104, 2005.
• D.D. Lee and H.S. Seung, Learning the parts of
  objects by nonnegative matrix factorization

24
Questions?