Inhalt

•  
   1 Introduction
•  
   1.1 Outline
•  
   1.2 Publications
•  
   2 Proteomics and mass spectrometry
•  
   2.1 Proteins and peptides: an overview
•  
   2.1.1 Post-translational modifications
•  
   2.2 Inside the machine -- mass spectrometry
•  
   2.2.1 Ion source
•  
   2.2.2 Mass analyzers
•  
   2.2.3 Detector
•  
   2.2.4 Tandem MS (MS/MS)
•  
   2.3 Protein separation techniques
•  
   2.4 LC coupled to ESI or MALDI -- principle and differences
•  
   2.5 Computational methods for proteomics
•  
   2.5.1 Identification of proteins with MS --- qualitative proteomics
•  
   2.5.2 Quantitative MS
•  
   2.5.3 Peak intensities
•  
   2.5.4 Relative and absolute quantification
•  
   3 Machine learning - methods and validation
•  
   3.1 Supervised learning
•  
   3.2 Linear regression
•  
   3.2.1 Properties
•  
   3.2.2 Implementations
•  
   3.3 Support vector machines
•  
   3.3.1 SVM for regression
•  
   3.3.2 Properties of SVR
•  
   3.3.3 Further reading and implementations
•  
   3.4 Random forests
•  
   3.4.1 Further reading and implementations
•  
   3.5 Shrinkage methods
•  
   3.6 Feature subset selection
•  
   3.6.1 Forward stepwise selection
•  
   3.6.2 Shrinkage methods for feature selection
•  
   3.6.3 Implementations
•  
   3.7 The two-sample t-test
•  
   3.8 Model evaluation
•  
   3.8.1 Structural risk minimization
•  
   3.8.2 Cross-validation
•  
   3.8.3 The Bayesian and Akaike information criteria
•  
   3.9 Pitfalls in statistical learning
•  
  4 Scope of this work
•  
   4.1 Related work
•  
   4.2 MS setups relevant to this work
•  
  5 Modeling the mass spectrometry process
•  
   5.1 The aim of the model
•  
   5.2 Steps and techniques
•  
   5.3 Sources of noise and errors
•  
   5.4 Accuracy enhancement of absolute quantitation with predicted peak intensities
•  
   6 Data acquisition, processing, and analysis
•  
  6.1 MALDI-TOF data
•  
   6.1.1 Wet lab procedures
•  
   6.1.2 In silico preprocessing
•  
   6.1.3 Construction of data sets
•  
   6.1.4 Normalization for MALDI datasets
•  
   6.1.5 Statistical analysis
•  
  6.2 LC-ESI data
•  
   6.2.1 Wet lab procedures
•  
   6.2.2 In silico preprocessing
•  
   6.2.3 Dataset construction
•  
   6.2.4 Normalization for LC-ESI data
•  
   6.2.5 Statistical analysis
•  
   6.3 Summary
•  
  7 Peak intensity prediction
•  
  7.1 Representation of Peptides
•  
   7.1.1 Computer scientist's paradigm: Peptides are strings
•  
   7.1.2 Biochemist's paradigm: Peptides are molecules
•  
  7.2 Statistical properties of the feature spaces
•  
   7.2.1 Correlated features
•  
   7.2.2 Frequency of dimers and trimers
•  
  7.3 Prediction
•  
   7.3.1 Methods
•  
   7.3.2 MALDI dataset prediction results
•  
   7.3.3 Results for LC-ESI data
•  
   7.3.4 Summary
•  
  8 Feature selection
•  
   8.1 Methods
•  
   8.2 Integration of selected features from different methods
•  
   8.2.1 Evaluation of the feature selection comparison for MALDI datasets
•  
  8.3 Detailed results of the different methods
•  
   8.3.1 Forward stepwise selection
•  
   8.3.2 Random forests for feature importance assessment
•  
   8.3.3 L1-penalized methods for feature selection
•  
   Feature selection by least-angle regression
•  
   Feature selection by L1-penalized generalized linear models
•  
   8.3.4 t-test in the seq feature set
•  
   8.4 Summary
•  
   9 Extended analysis
•  
  9.1 Detailed results
•  
   9.1.1 Analysis of error behavior
•  
   9.1.2 Are the predicted intensities the true signal?
•  
   9.1.3 Duplicate peptides in training and test set
•  
   9.1.4 Linear model
•  
   9.1.5 Unlabeled data
•  
   10 Conclusion
•  
  A Additional information
•  
   A.1 Notations
•  
   A.2 Abbreviations used for dataset variants:
•  
   A.3 Implementation details
•  
   A.4 The official IUPAC amino acid codes.
•  
   A.5 Overview of prediction results obtained in this work
•  
   B Glossary