SVM-based techniques for biomarker discovery in proteomic pattern data

1
SVM-based techniques for biomarker discovery in
proteomic pattern data

• Elena Marchiori
• Department of Computer Science
• Vrije Universiteit Amsterdam

2
Overview

• Variable selection
• SVM-based techniques
• Application to proteomic pattern data
• Results
• Conclusion

3
Variable Selection

• Select a small subset of input variables (for
  example genes in gene expression data, m/z values
  in proteomic pattern data) which are used for
  building classifier
• Advantages
• it is cheaper to measure less variables
• the resulting classifier is simpler and
  potentially faster
• prediction accuracy may improve by discarding
  irrelevant variables
• identifying relevant variables gives more insight
  into the nature of the corresponding
  classification problem (biomarker detection)

4
Support Vector Machines

• Advantages
• maximize the margin between two classes in the
  feature space characterized by a kernel function
• are robust with respect to high input dimension
• Disadvantages
• difficult to incorporate background knowledge
• Sensitive to outliers

5
Binary classification
f(x) sign(wTx b)
6
Linear Separators
7
SVM separable classes
Support vectors uniquely characterize optimal
hyper-plane
margin
Optimal hyper-plane
Support vector
8
SVM and outliers
outlier
9
SVM-RFE

• Linear binary classifier decision function
• Recursive Feature Elimination (SVM-RFE)
• at each iteration
• eliminate threshold of variables with lower
  score
• recompute scores of remaining variables
• SVM-RFE based algorithms
• run SVM-RFE with different thresholds
• JOIN select variables occurring more than cutoff
  times
• ENSEMBLE consider majority vote of resulting
  classifiers

10
SVM-RFE I. Guyon et al., Machine
Learning, 46,389-422, 2002
11
SVM-RFE variant

• Input Train set, threshold T, number N of
  variables to be selected
• Output subset of variables of size N
• RFE
• Train Run linear SVM on train set
• Score generate a sequence of variables ordered
  wrt the absolute value of their weight
• Eliminate remove T of variables from ordered
  sequence
• Repeat (train, score, eliminate) on train set
  restricted to remaining variables until only N
  variables are left

12
JOIN and ENSEMBLE SVM-RFE
13
Case Study proteomic pattern data

• Petricoin et al papers
• Commercial analysis software (Proteome Quest)
  http//www.correlogic.com/
• Data sets available at http//ncifdaproteomics.co
  m/ppatterns.php

14
Data generation SELDI-TOF MSSurface-enhanced
laser desorption/ionization time-of-flight mass
spectrometry

• Method for profiling a population of proteins in
  a sample according to the size and net charge of
  individual proteins.
• The readout is a spectrum of peaks. The position
  of a protein in the spectrum corresponds to its
  time of flight because the small proteins fly
  faster than the heavy ones.

1 Serum on protein binding plate 2 Insert plate
in vacuum chamber 3 Irradiate plate with laser 4
This launches the proteins / peptides 5 Measure
time of flight (TOF) of Ions, related to the
molecular weight of proteins
15
Example of proteomic pattern profile from one
blood sample
Abundance
Time of flight

• Heavier peptides move slower -gt
• Time of flight corresponds to weight
• Weight corresponds to peptides
• Measurement of relative abundance of detected
  peptides in serum

16
How to use such data?

• Diagnostic tool
• design a classifier for discriminating healthy
  from disease samples
• Biomarkers identification
• Variable subset selection (VSS) select a subset
  of input variables (m/z values) that best
  discriminate the two classes (potential
  biomarkers)

17
Commercial Tools

• Proteome Quest (Correlogic) GAclustering, no
  pre-selection (Petricoin et al., The Lancet 2002)
• Propeak (3Z Informatics) separability analysis
  bootstrap
• Biomarker AMplification Filter BAMF (Eclipse
  Diagnostics) ?

18
Non-commercial Techniques

• Pre-processing ranking kNN (Zhu et al., PNAS
  2003)
• Pre-selection boosted decision trees (Qu et
  al., Clin. Chem. 2002)
• Filter FS classifier (Liu et al., Genome
  Informatics 2002)
• GA SVM, SVM-RFE ensemble (Jong et al., EvoBIO
  2004, Jong et al. CIBCB 2004)
• Many others any ML method for classification/FS
  (see, e.g., special issue on FS, JMLR 2003)

19
Goal and Methods

• Goal analyze performance of SVM-based techniques
  for classification and variable selection with
  proteomic pattern data
• SVM
• SVM-RFE
• Ensemble SVM-RFE
• Majority vote of SVM-RFE classifiers obtained
  from SVM-RFE with different cutoff values
• Join SVM-RFE
• SVM trained on N variables that have been
  selected more often by SVM-RFE with different
  threshold values

20
DataSets

• Two proteomic pattern datasets from prostate and
  ovarian cancer from NCI/CCR and FDA/CBER Clinical
  proteomics Program Databank

M/z values
healthy
cancer
tot
15154
253
69
322
Prostate
15154
115 (15 benign)
100
Ovarian 4/03/02
215
Data sets available at http//ncifdaproteomics.c
om/ppatterns.php
21
Experimental Setup

• 10 random partitions of datasetT (50),H (25),V
  (25)
• Algorithms
• SVM trained on union of T and H
• SVM-RFE(threshold) with thresholds
  0.2,0.3,0.4,0.5, 0.6,0.7
• Choose threshold giving best classifier
  sensitivity on H
• JOIN(cutoff, 0.2, 0.3,0.4, 0.5,0.6,0.7) with
  cutoffs 1, 2, 3, 4, 5
• Choose cutoff giving best classifier sensitivity
  on H
• Performance average (over 10 V's)

22
Results Prostate Dataset
23
Results Ovarian Dataset
24
Controversy

• Noise, bias, results reliability and
  reproducibility in serum proteomics
• Sorace, Zhan, BMC Bioinformatics, 2004,
• Petricoin, BMC Bioinformatics, 2004,
• Baggerly, Journal of the National Cancer
  Institute, vol. 97, No.4, 2005.
• Liotta, Journal of the National Cancer Institute,
  vol. 97, No.4, 2005.
• Ransohoff, Journal of the National Cancer
  Institute, vol. 97, No.4, 2005.

25
Conclusion

• Many machine learning techniques can be used for
  potential biomarker detection with pattern
  proteomic data.
• SVM based techniques are a possible effective
  choice because of the high input dimension of
  such data.
• Computational analysis of pattern proteomic data
  has to use a correct methodology that considers
  biases induced by the selection and
  classification algorithms and by the data
  splitting.
• Problems related to reliability and
  reproducibility of data are inherent to the
  laboratory technology and actually addressed by
  researchers and practitioners.

26
Acknowledgments

• Connie Jimenez (Biology, VUMC)
• Aad van der Vaart (Statistics, VUA)