Harness the power of machine learning for quick and efficient calculations of protein structures and properties
Machine Learning in Protein Science is a unique and practical reference that shows how to employ machine learning approaches for full quantum mechanical (FQM) calculations of protein structures and properties, thereby saving costly computing time and making this technology available for routine users.
Machine Learning in Protein Science provides comprehensive coverage of topics including:
- Machine learning models and algorithms, from deep neural network (DNN) and transfer learning (TL) to hybrid unsupervised and supervised learning
- Protein structure predictions with AlphaFold to predict the effects of point mutations
- Modeling and optimization of the catalytic activity of enzymes
- Property calculations (energy, force field, stability, protein-protein interaction, thermostability, molecular dynamics)
- Protein design and large language models (LLMs) of protein systems
Machine Learning in Protein Science is an essential reference on the subject for biochemists, molecular biologists, theoretical chemists, biotechnologists, and medicinal chemists, as well as students in related programs of study.
Table of Contents:
1 Introduction 1
1.1 Background and Motivation 1
Bibliography 12
2 Fundamental of Theoretical Calculations on Protein Systems 15
2.1 Strategies for Protein Calculations and Predictions 15
2.2 Force Field-based Calculation on Protein Systems 21
2.3 Quantum Mechanical Calculation on Protein Systems 28
2.4 ml on Protein Systems 34
Bibliography 39
3 Protein Structure Prediction by Artificial Intelligence 43
3.1 AlphaFold from Google 43
3.1.1 Overview 43
3.1.2 Previous Work 43
3.1.3 What Is AlphaFold Expected to Do 44
3.1.4 Architecture of AlphaFold 44
3.1.4.1 Overall Architecture 44
3.1.4.2 Evoformer Block Architecture 45
3.1.4.3 Structure Prediction Model 47
3.1.5 Training Methods: Using Both Labeled and Unlabeled Data 48
3.1.6 Interpretation of the Network 48
3.1.7 Conclusion and Discussion 48
3.2 ESM and ESMFold from Meta AI 48
3.2.1 Meta’s ESM 2 49
3.2.1.1 Background 49
3.2.1.2 Methods 49
3.2.1.3 Evaluation and Results 49
3.2.1.4 Conclusion 51
3.2.2 Meta’s ESMFold 52
3.2.2.1 Related Work 52
3.2.2.2 Method 52
3.2.2.3 Results and Discussions 54
Bibliography 54
4 Methods and Tools for Predicting Protein Folding Free Energy Change upon Mutation 57
4.1 Introduction 57
4.1.1 Background 57
4.1.2 Experimental Methods 58
4.1.3 Non-machine Learning Methods 58
4.1.4 ML-based Methods 59
4.2 Clustered Tree Regression 60
4.2.1 Overview 61
4.2.2 Data Distribution and Cleaning 62
4.2.3 Feature Engineering 63
4.2.4 Model Training 65
4.2.5 Performance of CTR 65
4.3 Materials and Methods 68
4.3.1 Data Preparation 68
4.3.2 Feature Extraction 68
4.3.3 Experimental Setup 69
4.4 Conclusion 69
Bibliography 70
5 Deep Neural Network-assisted Full-system Quantum Mechanical (FQM) Calculations of Proteins 75
5.1 Introduction 75
5.1.1 Background 75
5.1.2 Advances in DNNs for Molecular Calculations 81
5.2 DNN-based GFCC Method 84
5.2.1 Methodology Overview 84
5.2.2 Methods 87
5.2.3 Results and Discussion 90
5.2.4 Conclusion 93
5.3 DNN-based Two-body Molecular Fractionation with Conjugate Caps 95
5.3.1 Methodology Overview 95
5.3.2 Neural Network Training and Energy Reconstruction 98
5.3.3 Discussion and Implications 101
5.4 Conclusion 106
Bibliography 107
6 Transfer Learning-assisted FQM Calculations of Proteins 111
6.1 Introduction 111
6.1.1 Background 111
6.1.2 Bridging the Gap: from Classical Methods to TL in Protein FQM Calculations 112
6.2 Inductive Transfer Learning Force Field 113
6.2.1 Introduction 113
6.2.2 Methods 114
6.2.3 Discussion and Implications 116
6.2.4 Conclusion and Outlook 118
6.3 Transfer-learning-based Deep Learning Protocol for FQM Calculations 120
6.3.1 Introduction 120
6.3.2 Methods 121
6.3.3 Discussion and Implications 124
6.3.4 Conclusion and Outlook 126
Bibliography 127
7 Protein Interaction Prediction with Artificial Intelligence 129
7.1 Background 129
7.2 Methods and Technical Framework 131
7.2.1 Sequence-based Methods 131
7.2.2 Structure-based Methods 133
7.2.3 Network-based Approaches 135
7.2.4 Text Mining and Literature-based Prediction 136
7.2.5 Hybrid Approaches 139
7.3 Results 141
7.4 Summary and Outlook 143
7.4.1 Summary of Achievements 143
7.4.2 Outlook and Future Directions 144
Bibliography 146
8 Protein Function Annotation with Machine Learning 149
8.1 Background 149
8.2 Methods and Technical Framework 152
8.3 Protein Function Annotation 154
8.3.1 Homology-based Annotation 156
8.3.2 Structural Annotation 156
8.3.3 Ontology-based Annotation 158
8.3.4 Network-based Annotation 159
8.3.5 ML- and AI-based Methods 160
8.3.6 High-throughput Experimental Techniques 161
8.3.7 Hybrid Methods 162
Bibliography 163
9 Machine Learning-driven Ab Initio Protein Design 165
9.1 Background 165
9.2 Advances and Applications in ML-driven Ab Initio Protein Design 167
9.3 Graph Neural Networks for Flexible Protein Design 171
9.3.1 Energy Function of Neural Networks for Protein Design 173
9.3.2 ml Models and Technologies 176
9.3.3 ml and AI-based Methods 177
9.3.4 Conclusion and Outlook 179
Bibliography 183
10 Large Language Model of Protein Systems 185
10.1 Background 185
10.2 Methodology and Training of Protein Language Models 189
10.2.1 Methodological Framework for Protein Language Models 189
10.2.2 Training Strategies for PLMs 191
10.3 Applications of PLMs 192
10.3.1 Case Studies in the Application of PLMs 192
10.4 Case Introduction: Decoding Protein Interactions Using ESM 195
10.4.1 Technical Foundations 195
10.4.2 Applications in Protein Design 196
10.4.3 Future Directions 198
10.5 Conclusion and Outlook 201
Bibliography 203
11 Outlook 207
11.1 Background 207
11.2 AI’s Transformative Role 210
11.3 Applications Across Protein Science 214
11.4 AI-Augmented Experimental Techniques in Modern Protein Research 217
Bibliography 220
Index 223
About the Author :
Jinjin Li is a Professor at the School of Electronic Information and Electrical Engineering, Shanghai Jiao Tong University in Shanghai, China. She performed postdoctoral work at the University of Illinois, USA and was a Senior Research Fellow at the University of California, USA.
Yanqiang Han is an Assistant Professor at the School of Electronic Information and Electrical Engineering, Shanghai Jiao Tong University in Shanghai, China.
