Statistical Diagnostics for Cancer Analyzing High-Dimensional Data

by Dehmer, Matthias; Emmert-Streib, Frank

• ISBN13: 9783527332625
• ISBN10: 3527332626
• Edition: 1st
• Format: Hardcover
• Copyright: 2013-02-25
• Publisher: Wiley-Blackwell

Summary

This ready reference discusses different methods for statistically analyzing and validating data created with high-throughput methods. As opposed to other titles, this book focusses on systems approaches, meaning that no single gene or protein forms the basis of the analysis but rather a more or less complex biological network. From a methodological point of view, the well balanced contributions describe a variety of modern supervised and unsupervised statistical methods applied to various large-scale datasets from genomics and genetics experiments. Furthermore, since the availability of sufficient computer power in recent years has shifted attention from parametric to nonparametric methods, the methods presented here make use of such computer-intensive approaches as Bootstrap, Markov Chain Monte Carlo or general resampling methods. Finally, due to the large amount of information available in public databases, a chapter on Bayesian methods is included, which also provides a systematic means to integrate this information. A welcome guide for mathematicians and the medical and basic research communities.

Author Biography

Frank Emmert-Streib studied physics at the University of Siegen (Germany) and received his Ph.D. in Theoretical Physics from the University of Bremen (Germany). He was a postdoctoral research associate at the Stowers Institute for Medical Research (Kansas City, USA) in the Department for Bioinformatics and a Senior Fellow at the University of Washington (Seattle, USA) in the Department of Biostatistics and the Department of Genome Sciences. Currently, he is Lecturer/Assistant Professor at the Queen's University Belfast at the Center for Cancer Research and Cell Biology (CCRCB) leading the Computational Biology and Machine Learning Lab. His research interests are in the field of computational biology, machine learning and biostatistics in the development and application of methods from statistics and machine learning for the analysis of high-throughput data from genomics and genetics experiments.

Matthias Dehmer studied mathematics at the University of Siegen (Germany) and received his PhD in computer science from the Technical University of Darmstadt (Germany). Afterwards, he was a research fellow at Vienna Bio Center (Austria), Vienna University of Technology and University of Coimbra (Portugal). Currently, he is Professor at UMIT - The Health and Life Sciences University (Austria). His research interests are in bioinformatics, cancer analysis, chemical graph theory, systems biology, complex networks, complexity, statistics and information theory. In particular, he is also working on machine learning-based methods to design new data analysis methods for solving problems in computational biology and medicinal chemistry.

Table of Contents

Preface XIII

List of Contributors XVII

Part One General Overview 1

1 Control of Type I Error Rates for Oncology Biomarker Discovery with High-Throughput Platforms 3
Jeffrey Miecznikowski, Dan Wang, and Song Liu

1.1 Brief Summary 3

1.2 Introduction 3

1.3 High-Throughput Platforms 4

1.3.1 Gene Expression Arrays 5

1.3.2 RNA-Seq 5

1.3.3 DNA Methylation Arrays 6

1.3.4 Mass Spectrometry Platforms 6

1.3.5 aCGH Arrays 7

1.3.6 Preprocessing HT Platforms 7

1.4 Analysis of Experiments 8

1.4.1 Linear Regression 8

1.4.1.1 Simple Linear Regression 9

1.4.1.2 Multiple Regression 11

1.4.2 Logistic Regression (Y Discrete) 11

1.4.2.1 Multiple Logistic Regression 13

1.4.3 Survival Modeling 13

1.4.3.1 Kaplan–Meier Analysis 13

1.5 Multiple Testing Type I Errors 15

1.5.1 FWER, k-FWER Methods 17

1.5.1.1 Adjusted Bonferroni Method 17

1.5.1.2 Holm Procedure 17

1.5.1.3 Generalized Hochberg Procedure 18

1.5.1.4 Generalized9S idak Procedure 18

1.5.1.5 minP and maxT procedures 19

1.6 Discussion 19

1.7 Perspective 20

References 21

2 Overview of Public Cancer Databases, Resources, and Visualization Tools 27
Frank Emmert-Streib, Ricardo de Matos Simoes, Shailesh Tripathi, and Matthias Dehmer

2.1 Brief Overview 27

2.2 Introduction 27

2.3 Different Cancer Types are Genetically Related 28

2.4 Incidence and Mortality Rates of Cancer 29

2.5 Cancer and Disorder Databases 30

2.6 Visualization and Network-Based Analysis Tools 34

2.6.1 Web-Based Software 34

2.6.2 R-Based Packages 34

2.7 Conclusions 35

2.8 Perspective 37

References 37

Part Two Bayesian Methods 41

3 Discovery of Expression Signatures in Chronic Myeloid Leukemia by Bayesian Model Averaging 43
Ka Yee Yeung

3.1 Brief Introduction 43

3.2 Chronic Myeloid Leukemia (CML) 44

3.3 Variable Selection on Gene Expression Data 44

3.4 Bayesian Model Averaging (BMA) 46

3.4.1 The Iterative BMA Algorithm (iBMA) 47

3.4.2 Computational Assessment 48

3.5 Case Study: CML Progression Data 49

3.6 The Power of iBMA 50

3.7 Laboratory Validation 51

3.8 Conclusions 52

3.9 Perspective 53

3.10 Publicly Available Resources 54

References 54

4 Bayesian Ranking and Selection Methods in Microarray Studies 57
Hisashi Noma and Shigeyuki Matsui

4.1 Brief Summary 57

4.2 Introduction 57

4.3 Hierarchical Mixture Modeling and Empirical Bayes Estimation 59

4.4 Ranking and Selection Methods 60

4.4.1 Ranking Based on Effect Sizes 60

4.4.1.1 Posterior Mean (PM) 61

4.4.1.2 Rank Posterior Mean (RPM) 61

4.4.1.3 Tail-Area Posterior Probability (TPP) 62

4.4.2 Ranking Based on Selection Accuracy of Differential Genes 63

4.4.2.1 Posterior Probability of Differentially Expressed (PPDE) 63

4.4.2.2 Evaluating Selection Accuracy 64

4.5 Simulations 65

4.6 Application 67

4.7 Concluding Remarks 71

4.8 Perspective 72

4.9 Appendix: The EM Algorithm 72

References 73

5 Multiclass Classification via Bayesian Variable Selection with Gene Expression Data 75
Yang Aijun, Song Xinyuan, and Li Yunxian

5.1 Brief Summary 75

5.2 Introduction 75

5.3 Matrix Variate Distribution 77

5.4 Method 77

5.4.1 Model 77

5.4.2 Prior Specification 79

5.4.3 Computation 80

5.4.4 Classification 82

5.5 Real Data Analysis 83

5.5.1 Leukemia Data 83

5.5.2 Lymphoma Data 87

5.5.3 Computational Time 89

5.6 Discussion 89

5.7 Perspective 89

References 90

6 Semisupervised Methods for Analyzing High-dimensional Genomic Data 93
Devin C. Koestler

6.1 Brief Summary 93

6.2 Motivation 93

6.3 Existing Approaches 95

6.3.1 Fully Unsupervised Procedures 96

6.3.2 Fully Supervised Procedures 96

6.3.3 Semisupervised Procedures 97

6.3.3.1 Semisupervised Clustering 99

6.3.3.2 Semisupervised RPMM 100

6.3.3.3 Considerations Regarding Semisupervised Procedures 101

6.4 Data Application: Mesothelioma Cancer Data Set 102

6.4.1 Results: Mesothelioma Cancer Data Set 104

6.5 Perspective 105

References 106

Part Three Network-Based Approaches 107

7 Colorectal Cancer and Its Molecular Subsystems: Construction, Interpretation, and Validation 109
Vishal N. Patel and Mark R. Chance

7.1 Brief Summary 109

7.2 Colon Cancer: Etiology 109

7.3 Colon Cancer: Development 110

7.4 The Pathway Paradigm 111

7.5 Cancer Subtypes and Therapies 112

7.6 Molecular Subsystems: Introduction 113

7.7 Molecular Subsystems: Construction 113

7.7.1 Measurements 113

7.7.2 Manifolds 114

7.8 Molecular Subsystems: Interpretation 117

7.8.1 Examples 117

7.9 Molecular Subsystems: Validation 119

7.10 Worked Example: Label-Free Proteomics 120

7.10.1 Whole Protein-Level Significance 122

7.10.2 Peptide-Level Significance 122

7.10.3 Exon-Level Significance 125

7.10.4 Summarizing the Results 126

7.11 Conclusions 127

7.12 Perspective 128

References 129

8 Network Medicine: Disease Genes in Molecular Networks 133
Sreenivas Chavali and Kartiek Kanduri

8.1 Brief Summary 133

8.2 Introduction 133

8.3 Genetic Architecture of Human Diseases 134

8.4 Systems Properties of Disease Genes 136

8.4.1 Network Measures 136

8.4.2 Disease and Disease-Gene Networks 137

8.4.3 Disease Genes in Protein Interaction Networks 139

8.4.4 Identification of Disease Modules 143

8.5 Disease Gene Prioritization 145

8.5.1 Linkage Methods 145

8.5.2 Disease-Module-Based Methods 146

8.5.3 Diffusion-Based Methods 147

8.6 Conclusion 147

8.7 Perspectives 148

References 148

9 Inference of Gene Regulatory Networks in Breast and Ovarian Cancer by Integrating Different Genomic Data 153
Binhua Tang, Fei Gu, and Victor X. Jin

9.1 Brief Summary 153

9.2 Introduction 153

9.3 Theory and Contents of Gene Regulatory Network 154

9.3.1 Basic Theory of Gene Regulatory Network 154

9.3.2 Content of Gene Regulatory Network 155

9.3.2.1 Identify and Infer the Structure Properties and Regulatory Relationships of Gene Networks 155

9.3.2.2 Understand the Basic Rules of Gene Expression and Function 155

9.3.2.3 Discover the Transfer Rules of Genetic Information During Gene Expression 155

9.3.2.4 Study on the Gene Function in a Systematic Framework 156

9.4 Inference of Gene Regulatory Networks in Human Cancer 156

9.4.1 The In Silico Analytical Approach 156

9.4.1.1 Study Case 1: Inference of Static Gene Regulatory Network of Estrogen-Dependent Breast Cancer Cell Line 158

9.4.1.2 Study Case 2: Gene Regulatory Network of Genome-Wide Mapping of TGFb/SMAD4 Targets in Ovarian Cancer Patients 160

9.4.2 A Bayesian Inference Approach for Genetic Regulatory Analysis 164

9.4.2.1 Study Case: ERa Transcriptional Regulatory Dynamics in Breast Cancer Cell 165

9.5 Conclusions 167

9.6 Perspective 168

References 169

10 Network-Module-Based Approaches in Cancer Data Analysis 173
Guanming Wu and Lincoln Stein

10.1 Brief Summary 173

10.2 Introduction 173

10.3 Notation and Terminology 174

10.4 Network Modules Containing Functionally Similar Genes or Proteins 174

10.5 Network Module Searching Methods 175

10.5.1 Greedy Network Module Search Algorithms 175

10.5.2 Objective Function Guided Search 176

10.5.3 Network Clustering Algorithms 176

10.5.4 Community Search Algorithms 177

10.5.5 Mutual Exclusivity-Based Search Algorithms 178

10.5.6 Weighted Gene Expression Network 178

10.6 Applications of Network-Module-Based Approaches in Cancer Studies 179

10.6.1 Network Modules and Cancer Prognostic Signatures 179

10.6.2 Cancer Driver Gene Search Based on Network Modules 179

10.6.3 Using Network Patterns to Identify Cancer Mechanisms 180

10.7 The Reactome FI Cytoscape Plug-in 180

10.7.1 Construction of a Functional Interaction Network 181

10.7.2 Network Clustering Algorithm 181

10.7.3 Cancer Gene Index Data Set 181

10.7.4 Analyzing the TCGA OV Mutation Data Set 182

10.7.4.1 Loading the Mutation File into Cytoscape and Constructing a FI Subnetwork 182

10.7.4.2 Network Clustering and Network Module Functional Analysis 184

10.7.4.3 Module-Based Survival Analysis 186

10.7.4.4 Cancer Gene Index Data Overlay Analysis 187

10.8 Conclusions 189

10.9 Perspective 189

References 191

11 Discriminant and Network Analysis to Study Origin of Cancer 193
Li Chen, Ye Tian, Guoqiang Yu, David J. Miller, Ie-Ming Shih, and Yue Wang

11.1 Brief Summary 193

11.2 Introduction 193

11.3 Overview of Relevant Machine Learning Techniques 194

11.3.1 Fisher’s Discriminant Analysis and ANOVA 194

11.3.2 Hierarchical Clustering 195

11.3.3 One-Versus-All Support Vector Machine and Nearest-Mean Classifier 196

11.3.4 Differential Dependency Network 197

11.4 Methods 198

11.4.1 CNA Data Analysis for Testing Existence of Monoclonality 198

11.4.1.1 Preprocessing 200

11.4.1.2 Assessing Statistical Significance of Monoclonality 200

11.4.1.3 Visualization of Monoclonality 201

11.4.2 A Two-Stage Analytical Method for Testing the Origin of Cancer 201

11.4.2.1 Basic Assumptions 202

11.4.2.2 Tissue Heterogeneity Correction 203

11.4.2.3 Stage 1: Feature Selection and Classification 203

11.4.2.4 Stage 2: Transcriptional Network Comparison 204

11.5 Experiments and Results 204

11.5.1 Monoclonality 204

11.5.1.1 Testing Existence of Monoclonality 204

11.5.1.2 The Significance of Monoclonality 206

11.5.2 Testing the Origin of Ovarian Cancer 207

11.5.2.1 Stage 1 Results 207

11.5.2.2 Stage 2 Results 208

11.6 Conclusion 211

11.7 Perspective 212

References 212

12 Intervention and Control of Gene Regulatory Networks: Theoretical Framework and Application to Human Melanoma Gene Regulation 215
Nidhal Bouaynaya, Roman Shterenberg, Dan Schonfeld, and Hassan M. Fathallah-Shaykh

12.1 Brief Summary 215

12.2 Gene Regulatory Network Models 216

12.3 Intervention in Gene Regulatory Networks 218

12.3.1 Optimal Stochastic Control 219

12.3.2 Heuristic Control Strategies 221

12.3.3 Structural Intervention Strategies 222

12.4 Optimal Perturbation Control of Gene Regulatory Networks 223

12.4.1 Feasibility Problem 226

12.4.2 Optimal Perturbation Control 226

12.4.2.1 Minimal-Energy Perturbation Control 226

12.4.2.2 Fastest-Convergence Rate Perturbation Control 228

12.4.3 Trade-offs Between Minimal-Energy and Fastest Convergence Rate Perturbation Control 228

12.4.4 Robustness of Optimal Perturbation Control 231

12.5 Human Melanoma Gene Regulatory Network 231

12.6 Perspective 235

References 236

Part Four Phenotype Influence of DNA Copy Number Aberrations 239

13 Identification of Recurrent DNA Copy Number Aberrations in Tumors 241
Vonn Walter, Andrew B. Nobel, D. Neil Hayes, and Fred A. Wright

13.1 Introduction 241

13.2 Genetic Background 242

13.2.1 Definitions 242

13.2.2 Mechanisms of DNA Copy Number Change: An Overview 243

13.2.3 CNAs and Cancer 244

13.2.4 Sporadic and Recurrent CNAs 245

13.2.5 Measuring DNA Copy Number 245

13.2.6 Other Issues to Consider When Assessing DNA Copy Number 246

13.3 Analyzing DNA Copy Number: Single Sample Methods 246

13.3.1 Notation 247

13.3.2 Quality Control and Preprocessing 247

13.3.3 Thresholding 247

13.3.4 Segmentation Algorithms 248

13.3.5 Methods Based on Hidden Markov Models 248

13.4 Analyzing DNA Copy Number Data: Multiple Sample Methods to Detect Recurrent CNAs 249

13.4.1 Additional Preprocessing and Summary Statistics 249

13.4.2 Multiple Testing 250

13.4.3 Assessing Statistical Significance: An Overview 250

13.5 Analyzing DNA Copy Number Data with DiNAMIC 251

13.5.1 Cyclic Shifts 251

13.5.2 Assessing Statistical Significance with DiNAMIC 252

13.5.3 Peeling 253

13.5.4 Confidence Intervals for Recurrent CNAs 256

13.5.5 Bootstrap Test-Based Confidence Intervals in Real Datasets 257

13.6 Open Questions 258

References 259

14 The Cancer Cell, Its Entropy, and High-Dimensional Molecular Data 261
Wessel N. van Wieringen and Aad W. van der Vaart

14.1 Brief Summary 261

14.2 Introduction 261

14.3 Background 262

14.3.1 Molecular Biology 262

14.3.2 Cancer 263

14.3.3 Measurement Devices 263

14.4 Entropy Increase 264

14.5 Statistical Arguments 266

14.6 Statistical Methodology 268

14.6.1 Experiments 269

14.6.2 Entropy 269

14.6.3 Mutual Information 272

14.7 Simulation 275

14.8 Application to Cancer Data 275

14.8.1 Analyses of Type II Experiments 276

14.8.2 Analyses of Type I Experiments 279

14.8.3 Potential 280

14.8.4 Discussion 282

14.9 Conclusion 283

14.10 Perspective 283

14.11 Software 284

References 284

Index 287