Batch Effects and Noise in Microarray Experiments

Andreas Scherer studied biology in Cologne, Germany, and Freiburg, Germany, and received his Ph.D. for his studies in the fields of genetics, developmental biology, and microbiology. Following a postdoctoral position at UT Southwestern Medical Center in Dallas, TX, he worked for many years in pharmaceutical industry in various positions in the field of experimental and statistical genomics biomarker discovery. In 2007, Andreas Scherer founded Spheromics, a company specialized in analytical and consultancy services in gene expression technologies and biomarker development.

• List of Contributors xiiiForeword xviiPreface xix1 Variation, Variability, Batches and Bias in Microarray Experiments: An Introduction 1Andreas Scherer2 Microarray Platforms and Aspects of Experimental Variation 5John A Coller Jr2.1 Introduction 52.2 Microarray Platforms 62.2.1 Affymetrix 62.2.2 Agilent 72.2.3 Illumina 72.2.4 Nimblegen 82.2.5 Spotted Microarrays 82.3 Experimental Considerations 92.3.1 Experimental Design 92.3.2 Sample and RNA Extraction 92.3.3 Amplification 122.3.4 Labeling 132.3.5 Hybridization 132.3.6 Washing 142.3.7 Scanning 152.3.8 Image Analysis and Data Extraction 162.3.9 Clinical Diagnosis 172.3.10 Interpretation of the Data 172.4 Conclusions 173 Experimental Design 19Peter Grass3.1 Introduction 193.2 Principles of Experimental Design 203.2.1 Definitions 203.2.2 Technical Variation 213.2.3 Biological Variation 213.2.4 Systematic Variation 223.2.5 Population, Random Sample, Experimental and Observational Units 223.2.6 Experimental Factors 223.2.7 Statistical Errors 233.3 Measures to Increase Precision and Accuracy 243.3.1 Randomization 253.3.2 Blocking 253.3.3 Replication 253.3.4 Further Measures to Optimize Study Design 263.4 Systematic Errors in Microarray Studies 283.4.1 Selection Bias 283.4.2 Observational Bias 283.4.3 Bias at Specimen/Tissue Collection 293.4.4 Bias at mRNA Extraction and Hybridization 303.5 Conclusion 304 Batches and Blocks, Sample Pools and Subsamples in the Design and Analysis of Gene Expression Studies 33Naomi Altman4.1 Introduction 334.1.1 Batch Effects 354.2 A Statistical Linear Mixed Effects Model for Microarray Experiments 354.2.1 Using the Linear Model for Design 374.2.2 Examples of Design Guided by the Linear Model 374.3 Blocks and Batches 394.3.1 Complete Block Designs 394.3.2 Incomplete Block Designs 394.3.3 Multiple Batch Effects 404.4 Reducing Batch Effects by Normalization and Statistical Adjustment 414.4.1 Between and Within Batch Normalization with Multi-array Methods 434.4.2 Statistical Adjustment 464.5 Sample Pooling and Sample Splitting 474.5.1 Sample Pooling 474.5.2 Sample Splitting: Technical Replicates 484.6 Pilot Experiments 494.7 Conclusions 49Acknowledgements 505 Aspects of Technical Bias 51Martin Schumacher, Frank Staedtler, Wendell D Jones, and Andreas Scherer5.1 Introduction 515.2 Observational Studies 525.2.1 Same Protocol, Different Times of Processing 525.2.2 Same Protocol, Different Sites (Study 1) 535.2.3 Same Protocol, Different Sites (Study 2) 555.2.4 Batch Effect Characteristics at the Probe Level 575.3 Conclusion 606 Bioinformatic Strategies for cDNA-Microarray Data Processing 61Jessica Fahlén, Mattias Landfors, Eva Freyhult, Max Bylesjö, Johan Trygg, Torgeir R Hvidsten, and Patrik Rydén6.1 Introduction 616.1.1 Spike-in Experiments 626.1.2 Key Measures – Sensitivity and Bias 636.1.3 The IC Curve and MA Plot 636.2 Pre-processing 646.2.1 Scanning Procedures 656.2.2 Background Correction 656.2.3 Saturation 676.2.4 Normalization 686.2.5 Filtering 706.3 Downstream Analysis 716.3.1 Gene Selection 716.3.2 Cluster Analysis 716.4 Conclusion 737 Batch Effect Estimation of Microarray Platforms with Analysis of Variance 75Nysia I George and James J Chen7.1 Introduction 757.1.1 Microarray Gene Expression Data 767.1.2 Analysis of Variance in Gene Expression Data 777.2 Variance Component Analysis across Microarray Platforms 787.3 Methodology 787.3.1 Data Description 787.3.2 Normalization 797.3.3 Gene-Specific ANOVA Model 817.4 Application: The MAQC Project 817.5 Discussion and Conclusion 85Acknowledgements 858 Variance due to Smooth Bias in Rat Liver and Kidney Baseline Gene Expression in a Large Multi-laboratory Data Set 87Michael J Boedigheimer, Jeff W Chou, J Christopher Corton, Jennifer Fostel, Raegan O’Lone, P Scott Pine, John Quackenbush, Karol L Thompson, and Russell D Wolfinger8.1 Introduction 878.2 Methodology 898.3 Results 898.3.1 Assessment of Smooth Bias in Baseline Expression Data Sets 898.3.2 Relationship between Smooth Bias and Signal Detection 918.3.3 Effect of Smooth Bias Correction on Principal Components Analysis 928.3.4 Effect of Smooth Bias Correction on Estimates of Attributable Variability 948.3.5 Effect of Smooth Bias Correction on Detection of Genes Differentially Expressed by Fasting 958.3.6 Effect of Smooth Bias Correction on the Detection of Strain-Selective Gene Expression 968.4 Discussion 97Acknowledgements 999 Microarray Gene Expression: The Effects of Varying Certain Measurement Conditions 101Walter Liggett, Jean Lozach, Anne Bergstrom Lucas, Ron L Peterson, Marc L Salit, Danielle Thierry-Mieg, Jean Thierry-Mieg, and Russell D Wolfinger9.1 Introduction 1019.2 Input Mass Effect on the Amount of Normalization Applied 1039.3 Probe-by-Probe Modeling of the Input Mass Effect 1039.4 Further Evidence of Batch Effects 1089.5 Conclusions 11010 Adjusting Batch Effects in Microarray Experiments with Small Sample Size Using Empirical Bayes Methods 113W Evan Johnson and Cheng li10.1 Introduction 11310.1.1 Bayesian and Empirical Bayes Applications in Microarrays 11410.2 Existing Methods for Adjusting Batch Effect 11510.2.1 Microarray Data Normalization 11510.2.2 Batch Effect Adjustment Methods for Large Sample Size 11510.2.3 Model-Based Location and Scale Adjustments 11610.3 Empirical Bayes Method for Adjusting Batch Effect 11710.3.1 Parametric Shrinkage Adjustment 11710.3.2 Empirical Bayes Batch Effect Parameter Estimates using Nonparametric Empirical Priors 12010.4 Data Examples, Results and Robustness of the Empirical Bayes Method 12110.4.1 Microarray Data with Batch Effects 12110.4.2 Results for Data Set 1 12410.4.3 Results for Data Set 2 12410.4.4 Robustness of the Empirical Bayes Method 12610.4.5 Software Implementation 12710.5 Discussion 12811 Identical Reference Samples and Empirical Bayes Method for Cross-Batch Gene Expression Analysis 131Wynn L Walker and Frank R Sharp11.1 Introduction 13111.2 Methodology 13311.2.1 Data Description 13311.2.2 Empirical Bayes Method for Batch Adjustment 13411.2.3 Naïve t-test Batch Adjustment 13511.3 Application: Expression Profiling of Blood from Muscular Dystrophy Patients 13511.3.1 Removal of Cross-Experimental Batch Effects 13511.3.2 Removal of Within-Experimental Batch Effects 13611.3.3 Removal of Batch Effects: Empirical Bayes Method versus t-Test Filter 13711.4 Discussion and Conclusion 13811.4.1 Methods for Batch Adjustment Within and Across Experiments 13811.4.2 Bayesian Approach is Well Suited for Modeling Cross-Experimental Batch Effects 13911.4.3 Implications of Cross-Experimental Batch Corrections for Clinical Studies 13912 Principal Variance Components Analysis: Estimating Batch Effects in Microarray Gene Expression Data 141Jianying Li, Pierre R Bushel, Tzu-Ming Chu, and Russell D Wolfinger12.1 Introduction 14112.2 Methods 14312.2.1 Principal Components Analysis 14312.2.2 Variance Components Analysis and Mixed Models 14512.2.3 Principal Variance Components Analysis 14512.3 Experimental Data 14612.3.1 A Transcription Inhibition Study 14612.3.2 A Lung Cancer Toxicity Study 14712.3.3 A Hepato-toxicant Toxicity Study 14712.4 Application of the PVCA Procedure to the Three Example Data Sets 14812.4.1 PVCA Provides Detailed Estimates of Batch Effects 14812.4.2 Visualizing the Sources of Batch Effects 14912.4.3 Selecting the Principal Components in the Modeling 15012.5 Discussion 15313 Batch Profile Estimation, Correction, and Scoring 155Tzu-Ming Chu, Wenjun Bao, Russell S Thomas, and Russell D Wolfinger13.1 Introduction 15513.2 Mouse Lung Tumorigenicity Data Set with Batch Effects 15713.2.1 Batch Profile Estimation 15913.2.2 Batch Profile Correction 16013.2.3 Batch Profile Scoring 16113.2.4 Cross-Validation Results 16213.3 Discussion 164Acknowledgements 16514 Visualization of Cross-Platform Microarray Normalization 167Xuxin Liu, Joel Parker, Cheng Fan, Charles M Perou, and J S Marron14.1 Introduction 16714.2 Analysis of the NCI 60 Data 16914.3 Improved Statistical Power 17414.4 Gene-by-Gene versus Multivariate Views 17814.5 Conclusion 18115 Toward Integration of Biological Noise: Aggregation Effect in Microarray Data Analysis 183Lev Klebanov and Andreas Scherer15.1 Introduction 18315.2 Aggregated Expression Intensities 18515.3 Covariance between Log-Expressions 18615.4 Conclusion 189Acknowledgements 19016 Potential Sources of Spurious Associations and Batch Effects in Genome-Wide Association Studies 191Huixiao Hong, Leming Shi, James C Fuscoe, Federico Goodsaid, Donna Mendrick, and Weida Tong16.1 Introduction 19116.2 Potential Sources of Spurious Associations 19216.2.1 Spurious Associations Related to Study Design 19416.2.2 Spurious Associations Caused in Genotyping Experiments 19516.2.3 Spurious Associations Caused by Genotype Calling Errors 19516.3 Batch Effects 19616.3.1 Batch Effect in Genotyping Experiment 19616.3.2 Batch Effect in Genotype Calling 19716.4 Conclusion 201Disclaimer 20117 Standard Operating Procedures in Clinical Gene Expression Biomarker Panel Development 203Khurram Shahzad, Anshu Sinha, Farhana Latif, and Mario C Deng17.1 Introduction 20317.2 Theoretical Framework 20417.3 Systems-Biological Concepts in Medicine 20417.4 General Conceptual Challenges 20517.5 Strategies for Gene Expression Biomarker Development 20517.5.1 Phase 1: Clinical Phenotype Consensus Definition 20617.5.2 Phase 2: Gene Discovery 20717.5.3 Phase 3: Internal Differential Gene List Confirmation 20917.5.4 Phase 4: Diagnostic Classifier Development 20917.5.5 Phase 5: External Clinical Validation 21017.5.6 Phase 6: Clinical Implementation 21117.5.7 Phase 7: Post-Clinical Implementation Studies 21217.6 Conclusions 21318 Data, Analysis, and Standardization 215Gabriella Rustici, Andreas Scherer, and John Quackenbush18.1 Introduction 21518.2 Reporting Standards 21618.3 Computational Standards: From Microarray to Omic Sciences 21918.3.1 The Microarray Gene Expression Data Society 21918.3.2 The Proteomics Standards Initiative 22018.3.3 The Metabolomics Standards Initiative 22018.3.4 The Genomic Standards Consortium 22018.3.5 Systems Biology Initiatives 22118.3.6 Data Standards in Biopharmaceutical and Clinical Research 22118.3.7 Standards Integration Initiatives 22218.3.8 The MIBBI project 22318.3.9 OBO Foundry 22318.3.10 FuGE and ISA-TAB 22318.4 Experimental Standards: Developing Quality Metrics and a Consensus on Data Analysis Methods 22618.5 Conclusions and Future Perspective 228References 231Index 245