Statistical Modelling of Molecular Descriptors in QSAR/QSPR.

Statistical Modelling of Molecular Descriptors in QSAR/QSPR -- Contents -- Preface -- List of Contributors -- 1 Current Modeling Methods Used in QSAR/QSPR -- 1.1 Introduction -- 1.2 Modeling Methods -- 1.2.1 Methods for Regression Problems -- 1.2.1.1 Multiple Linear Regression -- 1.2.1.2 Partial Least Squares -- 1.2.1.3 Feedforward Backpropagation Neural Network -- 1.2.1.4 General Regression Neural Network -- 1.2.1.5 Gaussian Processes -- 1.2.2 Methods for Classification Problems -- 1.2.2.1 Logistic Regression -- 1.2.2.2 Linear Discriminant Analysis -- 1.2.2.3 Decision Tree and Random Forest -- 1.2.2.4 k-Nearest Neighbor -- 1.2.2.5 Probabilistic Neural Network -- 1.2.2.6 Support Vector Machine -- 1.3 Software for QSAR Development -- 1.3.1 Structure Drawing or File Conversion -- 1.3.2 3D Structure Generation -- 1.3.3 Descriptor Calculation -- 1.3.4 Modeling -- 1.3.5 General purpose -- 1.4 Conclusion -- References -- 2 Developing Best Practices for Descriptor-Based Property Prediction: Appropriate Matching of Datasets, Descriptors, Methods, and Expectations -- 2.1 Introduction -- 2.1.1 Posing the Question -- 2.1.2 Validating the Models -- 2.1.3 Interpreting the Models -- 2.2 Leveraging Experimental Data and Understanding their Limitations -- 2.3 Descriptors: The Lexicon of QSARs -- 2.3.1 Classical QSAR Descriptors and Uses -- 2.3.2 Experimentally Derived Descriptors -- 2.3.2.1 Biodescriptors -- 2.3.2.2 Descriptors from Spectroscopy/Spectrometry and Microscopy -- 2.3.3 0D, 1D and 2D Computational Descriptors -- 2.3.4 3D Descriptors and Beyond -- 2.3.5 Local Molecular Surface Property Descriptors -- 2.3.6 Quantum Chemical Descriptors -- 2.4 Machine Learning Methods: The Grammar of QSARs -- 2.4.1 Principal Component Analysis -- 2.4.2 Factor Analysis.

2.4.3 Multidimensional Scaling, Stochastic Proximity Embedding, and Other Nonlinear Dimensionality Reduction Methods -- 2.4.4 Clustering -- 2.4.5 Partial Least Squares (PLS) -- 2.4.6 k-Nearest Neighbors (kNN) -- 2.4.7 Neural Networks -- 2.4.8 Ensemble Models -- 2.4.9 Decision Trees and Random Forests -- 2.4.10 Kernel Methods -- 2.4.11 Ranking Methods -- 2.5 Defining Modeling Strategies: Putting It All Together -- 2.6 Conclusions -- References -- 3 Mold2 Molecular Descriptors for QSAR -- 3.1 Background -- 3.1.1 History of QSAR -- 3.1.2 Introduction to QSAR -- 3.1.3 Molecular Descriptors: Bridge for QSAR -- 3.1.3.1 Molecular Descriptors -- 3.1.3.2 Role of Molecular Descriptors -- 3.1.3.3 Types of Molecular Descriptors -- 3.1.3.4 Calculation of Molecular Descriptors (Software Packages) -- 3.2 Mold2 Molecular Descriptors -- 3.2.1 Description of Mold2 Descriptors -- 3.2.1.1 Topological Descriptors -- 3.2.1.2 Constitutional Descriptors -- 3.2.1.3 Information Content-based Descriptors -- 3.2.2 Calculation of Mold2 Descriptors -- 3.2.3 Evaluation of Mold2 Descriptors -- 3.2.3.1 Information Content by Shannon Entropy Analysis -- 3.2.3.2 Correlations between Descriptors -- 3.3 QSAR Using Mold2 Descriptors -- 3.3.1 Classification Models based on Mold2 Descriptors -- 3.3.2 Regression Models based on Mold2 Descriptors -- 3.4 Conclusion Remarks -- References -- 4 Multivariate Analysis of Molecular Descriptors -- 4.1 Introduction -- 4.2 2D Matrix-Based Descriptors -- 4.3 Graph-Theoretical Matrices -- 4.3.1 Vertex Weighting Schemes -- 4.4 Multivariate Similarity Analysis of Chemical Spaces -- 4.5 Analysis of Chemical Information of Descriptors from Graph-Theoretical Matrices -- 4.5.1 Data Sets -- 4.5.2 Comparison of Graph-Theoretical Matrices -- 4.5.2.1 Comparison of Weighted Graph-Theoretical Matrices -- 4.5.3 Comparison of Matrix Operators.

4.5.4 Comparison of Single Operators from Different Graph-Theoretical Matrices -- 4.6 Conclusions -- References -- 5 Partial-Order Ranking and Linear Modeling: Their Use in Predictive QSAR/QSPR Studies -- 5.1 Introduction -- 5.2 Linear QSAR Methodology, ERM, RM and GA -- 5.2.1 Replacement Method -- 5.2.2 Enhanced Replacement Method -- 5.2.3 Genetic Algorithm -- 5.2.4 Main Differences between MRM and RM -- 5.3 Principles of Ranking Methods -- 5.4 Selection of the Molecular Descriptors for Ranking -- 5.5 QSAR Based on Hasse Diagrams -- 5.6 Discussion -- 5.7 Conclusions -- References -- 6 Graph-Theoretical Descriptors for Branched Polymers -- 6.1 Introduction -- 6.2 Algebraic Graph Theory -- 6.3 Ideal Chain Models -- 6.4 Graph-Theoretical Approach to Chain Dynamics and Statistics -- 6.4.1 Radius of Gyration -- 6.4.2 Rouse Dynamics -- 6.4.3 Intrinsic Viscosity -- 6.4.4 Scattering Function -- 6.4.5 High Moments of Relaxation Time and Radius of Gyration -- 6.5 Applications -- 6.6 Final Remarks -- References -- 7 Structural-Similarity-Based Approaches for the Development of Clustering and QSPR/QSAR Models in Chemical Databases -- 7.1 Chemical Structural Similarity -- 7.1.1 Molecular Graph and Structural Similarity -- 7.1.2 Descriptor-Based Structural Similarity -- 7.1.3 Combining Structural Similarity Approaches -- 7.1.4 Approximate Structural Similarity -- 7.2 Clustering Models Based on Structural Similarity -- 7.2.1 Clustering of Chemical Databases -- 7.2.1.1 Pattern Representation of Chemicals Structures -- 7.2.1.2 Clustering of Chemical Databases -- 7.3 QSPR/QSAR Models Based on Structural Similarity -- 7.3.1 Dataset Selection -- 7.3.2 Dataset Representation -- 7.3.3 Fitting of the Dataset Representation -- 7.3.4 Building and Validation of the QSAR Model -- References.

8 Statistical Methods for Predicting Compound Recovery Rates for Ligand- Based Virtual Screening and Assessing the Probability of Activity -- 8.1 Introduction -- 8.2 Theory -- 8.2.1 Bayesian Approach to Virtual Screening -- 8.2.2 Predicting the Performance of Bayesian Screening -- 8.2.3 Practical Prediction of Compound Recall -- 8.2.4 Exemplary Results -- 8.3 Alternative Approaches to the Prediction of Compound Recall -- 8.4 Conclusions -- References -- 9 Molecular Descriptors and the Electronic Structure -- 9.1 Introduction -- 9.2 The Structure of Molecules -- 9.2.1 General Remarks -- 9.2.2 Structure Coding -- 9.2.3 Structural Features -- 9.2.4 Structure and Energy -- 9.3 The Electronic Structure -- 9.4 Dividing Molecules in Atoms and Bonds -- 9.4.1 Bonding in Molecules -- 9.4.2 Energy Partitioning -- 9.4.3 Energy and the Hückel Approach -- 9.4.4 Energy Components of Atoms and Bonds -- 9.4.5 Perturbation Treatment of the Electronic Structure -- 9.4.6 Thermodynamic Equilibrium -- 9.4.7 Model of ''Atom in Molecules'' -- 9.5 Structure and Dynamics -- 9.5.1 Molecular Flexibility -- 9.5.2 Molecular Dynamics Simulation -- 9.5.3 Conformational Space -- 9.6 Structure and Properties -- 9.6.1 Structure Property Relationships -- 9.6.2 Type of Molecular Properties -- 9.6.3 Molecular Commonality and Similarity -- 9.6.4 Multilinear Regression -- 9.6.5 Selection of Molecular Descriptors -- 9.7 Modeling of Physicochemical Properties of the Isomers of Hexane -- 9.8 Modeling of the Proton Affinity -- 9.8.1 Proton Affinity of Pyridines -- 9.8.1.1 Data and Mechanism -- 9.8.1.2 Model I -- 9.8.1.3 Model II -- 9.8.1.4 Model III -- 9.8.1.5 Model IV -- 9.8.1.6 Model V -- 9.8.1.7 Model VI -- 9.8.2 Basicity of N-Heterocyclic Aromatics -- 9.9 Molecular Surface Properties -- 9.10 Conclusions -- References -- 10 New Types of Descriptors and Models in QSAR/QSPR.

10.1 Introduction -- 10.2 Local Properties -- 10.2.1 Molecular Electrostatic Potential -- 10.2.2 Electron Density -- 10.2.3 Local Polarizability -- 10.2.4 Local Ionization Energy and Local Electron Affinity -- 10.3 Descriptors Derived from Local Properties -- 10.3.1 PEST Methodology -- 10.4 MEP as Descriptor for Hydrogen-Bonding Strengths -- 10.5 ParaSurf (Politzer-Murray) Descriptors -- 10.6 4D: Conformational-Ensemble-based Descriptors -- 10.7 Proper Validation/Generation of QSA(P)R Models -- 10.8 Conclusions -- References -- 11 Consensus Models of Activity Landscapes -- 11.1 Introduction -- 11.2 Characterization of the Activity Landscape -- 11.3 Consensus Models of Activity Landscape -- 11.3.1 Chemical Space and Molecular Representation -- 11.3.2 Activity Landscape with Multiple Representations -- 11.4 Conclusions and Future Perspectives -- References -- 12 Reverse Engineering Chemical Reaction Networks from Time Series Data -- 12.1 Introduction -- 12.2 Problem Definition -- 12.3 Reconstruction of Elementary Reaction Networks from Data by Network Search -- 12.3.1 Network Search as a Nonlinear Integer Programming Problem -- 12.3.2 Estimation of the Rate Coefficients for Trial Reaction Networks -- 12.4 Formulation of the Objective Function for Network Search -- 12.4.1 Physical/Chemical Information Available 336 -- 12.4.2 No physical/Chemical Information Available -- 12.5 Differential Evolution for Searching the Space of Reaction Networks -- 12.5.1 Basic DE Optimization Method -- 12.5.2 Self-Adaptive DE with Integer Variables -- 12.6 Network Identification Case Studies -- 12.6.1 Estimation of Time Derivatives -- 12.6.2 DE Settings -- 12.6.3 Model Selection Methodology -- 12.6.4 Results -- 12.7 Conclusions -- References.

13 Reduction of Dimensionality, Order, and Classification in Spaces of Theoretical Descriptions of Molecules: An Approach Based on Metrics, Pattern Recognition Techniques, and Graph Theoretic Considerations.