Machine Learning and Big Data-enabled Biotechnology

Dr. Hal S. Alper is the Cockrell Family Regents Chair in Engineering #1 at The University of Texas at Austin in the McKetta Department of Chemical Engineering. His research focuses on applying and extending the approaches of metabolic engineering, synthetic biology, systems biology, and protein engineering.

• Preface xv1 From Genome to Actionable Insights in Biotechnology 1James Morrissey, Benjamin Strain, and Cleo Kontoravdi1.1 Introduction 11.2 From Genome to Network 21.2.1 Metabolic Networks 31.2.1.1 Bottom-Up Approaches for Network Reconstruction 31.2.1.2 Top-Down Approaches for Network Reconstruction 41.2.2 Networks Beyond Metabolism 51.3 From Draft to Functional Network 61.3.1 Additional Reactions 61.3.1.1 Exchange Reactions 61.3.1.2 Demand Reactions 61.3.1.3 Transport Reactions 61.3.1.4 Spontaneous Reactions 71.3.1.5 Nongrowth Associated ATP Maintenance 71.3.1.6 Biomass Reaction 71.3.2 Network Validation 71.3.2.1 Manual Screening 81.3.2.2 Screening for Dead-End Reactions and Blocked Metabolites 81.3.2.3 Infinite Loops 91.3.2.4 Leaks and Siphons 101.4 From Functional Network to Model 101.4.1 Flux Balance Analysis 111.4.2 Flux Variability Analysis 121.4.3 Flux Sampling 131.5 From Model to In Silico Predictions 151.5.1 Constraints 151.5.2 Objective Function 161.5.3 Validating In Silico Predictions 161.5.3.1 Growth Rate Predictions 221.5.3.2 Amino Acid Auxotrophies 221.5.3.3 Gene Essentialities 221.5.3.4 Known Host Traits 231.5.3.5 Intracellular Predictive Accuracy 231.5.4 Toward Multilayer, Multiscale Metabolic Networks 231.5.4.1 Integrating Gene Regulatory Networks 241.5.4.2 Integrating Transcription and Translation 251.5.4.3 Integrating Signaling Networks 251.5.4.4 Multicellular and Multitissue Models 251.5.4.5 Multiscale Bioreactor Models 261.6 From Predictions to Actionable Insights in Biotechnology 261.6.1 Metabolic Engineering 261.6.2 Cell Line Development and Metabolic Profiling 271.6.3 Media and Feed Design 291.6.4 Gene Essentiality 301.6.5 Kinetic Parameter Estimation 301.6.6 Process Monitoring and Forecasting 30References 312 Automated Approaches for the Development of Genome-Scale Metabolic Network Models 43Emma M. Glass, Deborah A. Powers, and Jason A. Papin2.1 Introduction 432.2 Manual GSM Creation 442.3 Automated GSM Development 452.3.1 General Approach for Automated GSM Methods 452.3.2 GSM Construction Tools 462.3.2.1 From Raw Sequences 462.3.2.2 From Pre-annotated Sequences 522.3.2.3 From Reaction Database Information 532.3.2.4 Based on Existing GSMs 562.3.2.5 GSM Modification and Visualization Tools 572.4 Applications of Automatically-Generated GSM Collections 592.4.1 AGORA1 – 773 GSMs 592.4.2 EMBL GEMs – 5,587 GSMs 602.4.3 MetaGEM – 447 GSMs 612.4.4 AGORA2–7,302 GSMs 612.4.5 PATHGENN – 914 GSMs 622.5 Future Directions for the Field of Automated GSM Development 622.6 Conclusion 63References 633 Machine-Guided Approaches for Synthetic Biology Part Design 67Marc Amil, Leandro N. Ventimiglia, and Aleksej Zelezniak3.1 Introduction 673.2 Model-Guided Sequence Design Using Deep Learning 703.2.1 Predictive Models for DNA Function: CNNs in Regulatory Sequence Analysis 703.2.1.1 Data Considerations for Supervised Learning on Genomic Sequences 723.2.1.2 Primer on Convolutional Neural Networks for Supervised Genomic Sequence Modeling 733.2.2 Generative Sequence Modeling 753.2.2.1 Data Preparation for Unsupervised Learning 763.2.2.2 GANs for the Design of Biological Sequences 773.2.2.3 Transformer-Based DNA Models 803.2.2.4 Diffusion Models for the Design of Biological Sequences 843.3 Sources of Sequence–Function Data for Deep Learning 853.3.1 Native-Context Genome-Derived Datasets 863.3.2 Synthetic Datasets 873.4 Evaluating Synthetic Biological Parts Using Motif Analysis and Deep Learning 903.5 Current Challenges of Generative Part Design 933.6 Conclusion 93References 944 Machine Learning for Sequence-to-Function Approaches 103Rana A. Barghout, Maxim Kirby, Austin Zheng, Lya Chinas, Marjan Mohammadi, Zhiqing Xu, Benjamin Sanchez-Lengeling, and Radhakrishnan Mahadevan4.1 Introduction 1034.2 Current State of Sequence-to-Function Modeling 1054.2.1 Protein Function Prediction: From BLAST to Language Models 1054.2.2 Gene Ontology 1064.2.3 Enzyme Commission Numbers 1064.2.4 Enzyme Activity 1094.2.5 Protein Thermal Stability 1094.2.6 Protein Toxicity 1104.2.7 Protein Solubility 1114.3 Tool Kits and Benchmarks 1114.3.1 Overview of Open-Source Tools 1114.3.2 Importance of Standardized Benchmarks 1144.4 Emerging ML Methods 1154.4.1 Contrastive Learning 1154.4.2 Meta Learning 1164.5 Case Studies 1164.5.1 Prediction of Enzyme Activity and Substrate Specificity 1164.5.1.1 Predicting k cat Using CPI-Pred 1174.6 Challenges in Sequence-to-Function Mapping 1184.6.1 Sparse Experimental Data 1194.6.2 Interpretability 1204.7 Conclusion and Future Directions 120References 1215 Prediction of Enzyme Functions by Artificial Intelligence 131Ha Rim Kim, Hongkeun Ji, Gi Bae Kim, and Sang Yup Lee5.1 Introduction 1315.2 Conventional Computational Approaches for Predicting Enzyme Function 1325.3 Prediction of Enzyme Functions Using Machine Learning 1335.3.1 Extraction of Enzyme Features from Amino Acid Sequences 1345.3.2 Machine Learning-Based Approaches Algorithms for Enzyme Function Prediction 1365.4 Prediction of Enzyme Functions Using Deep Learning 1385.4.1 Convolutional Neural Network 1395.4.2 Recurrent Neural Network 1415.4.3 Transformer and Protein Language Models 1425.4.4 Graph Neural Network 1455.5 Concluding Remarks 147Acknowledgments 153References 1536 Design of Biochemical Pathways via AI/ML-Enabled Retrobiosynthesis 161Hongxiang Li, Xuan Liu, and Huimin Zhao6.1 Introduction 1616.1.1 Computer-Aided Synthesis Planning 1616.1.2 Retrobiosynthesis 1626.2 Retrobiosynthesis Tools 1626.2.1 Template-Based Tools 1626.2.2 Template-Free Tools 1666.2.3 Searching Algorithm 1676.2.4 Ranking 1686.3 Enzyme Selection and Optimization 1686.3.1 Enzyme Substrate Specificity 1696.3.2 Enzyme Catalytic Efficiency 1706.3.3 Enzyme Engineering 1726.3.4 De Novo Enzyme Design and Discovery 1726.4 Perspectives 1746.4.1 Integrating Biocatalysis with Chemocatalysis 1756.4.2 Toward Next-Generation AI for Retrobiosynthesis Planning 1756.4.3 Enhancing Enzyme Prediction and Design Capabilities 1766.4.4 Data Standardization and Model Interpretability 177Acknowledgments 178References 1797 Machine Learning to Accelerate the Discovery of Therapeutic Peptides 183Nicole Soto-Garcia, Mehdi D. Davari, and David Medina-Ortiz7.1 Introduction 1837.2 Peptides: Definitions and Main Characteristics 1847.3 Benefits and Limitations of Therapeutic Peptides 1857.4 Computational Design of Therapeutic Peptides 1867.5 Data Sources for Peptide Discovery 1877.6 ML-Based Strategies to Accelerate the Discovery of Therapeutic Peptides 1897.6.1 Data-Driven Approaches 1907.6.2 ml Strategies for Peptide Bioactivity Classification 1927.6.2.1 Classification Models for Antimicrobial Peptides 1927.6.2.2 Classification Models for Antiviral Peptides 1937.6.2.3 Classification Models for Antifungal Peptides 1947.6.2.4 Additional Classification Models for Therapeutic Peptides 1947.6.3 Strategies for Building Toxicity Classification Models 1957.6.3.1 Toxicity Prediction 1967.6.3.2 Immunogenicity Identification 1977.6.3.3 Hemolysis Evaluation 1977.6.3.4 Other Toxic Adverse Effect Predictions 1987.6.4 Data-Driven Strategies for Modeling Pharmacological Profiles 1987.6.5 De novo Design of Therapeutic Peptides 1997.6.5.1 Variational Autoencoder-Based Approaches 2007.6.5.2 Generative Adversarial Networks 2007.6.5.3 Transformer-Based Language Models 2017.6.5.4 Diffusion Models 2017.7 Next-Generation Peptide Design Through Multi-Agent Systems 2017.8 Developing AI-Agent for Autonomous Therapeutic Peptide Design 2037.9 Conclusion and Perspectives 205Acknowledgments 206References 2078 Machine Learning Approaches for High-Throughput Microbial Identification/Culturing 219Mohamed Mastouri and Yang Zhang8.1 Introduction 2198.2 High-Throughput (HTP) Techniques in Microbial Research 2218.2.1 Definition and Scope of HTP Microbial Techniques 2218.2.2 Metagenomics and Next-Generation Sequencing (NGS) 2238.2.3 Mass Spectrometry-Based Proteomics (MALDI-TOF MS) 2238.2.4 Flow Cytometry 2248.2.5 Microfluidics and Lab-on-a-Chip Systems 2248.2.6 High-Content Imaging and Phenotyping 2258.3 Fundamentals of Machine Learning 2268.3.1 Definition of Machine Learning and AI 2268.3.2 Supervised vs. Unsupervised vs. Reinforcement Learning 2268.3.3 ML Algorithms Commonly Used in Microbial Identification 2298.4 Machine Learning Approaches for High-Throughput Microbial Identification 2308.4.1 Genomic and Metagenomic Data Processing 2308.4.2 Mass Spectrometry-Based Identification 2348.4.3 Imaging-Based Identification 2368.5 Machine Learning Approaches for High-Throughput Microbial Culturing 2378.5.1 ML-Driven Microbial Growth Prediction 2378.5.2 AI in Microbial Cultivation Process Optimization 2388.5.3 Synthetic Biology and AI-Driven Strain Engineering 2408.6 Challenges and Limitations of Machine Learning in HTP Microbial Research 2418.7 Future Perspectives and Emerging Trends 2428.8 Conclusion 243Acknowledgments 244References 2449 Generative AI for Knowledge Mining of Synthetic Biology and Bioprocess Engineering Literature 253Zhengyang Xiao and Yinjie J. Tang9.1 Introduction 2539.2 Text Mining Using Knowledge Graph Tools 2549.2.1 NEKO: A Lightweight Knowledge Graph Tool 2549.2.2 GraphRAG 2569.3 LLM-Automated Data Extraction for Machine Learning 2589.4 Current Limitations 2599.5 Conclusion 260Acknowledgments 260References 26010 Metabolomics: Big Data Approaches 263Kenya Tanaka, Christopher J. Vavricka, and Tomohisa Hasunuma10.1 Introduction 26310.2 Methods for Metabolomics 26410.2.1 Preparation of Samples 26410.2.2 Detection and Quantification 26610.3 Analysis and Application of Metabolomics Data for Biotechnology 26710.3.1 Metabolomics for Identification of Pathway Bottlenecks 26710.3.2 Absolute Metabolomics and Thermodynamic Analyses 27210.3.3 Evaluation and Optimization of Metabolic Flux 27210.4 Artificial Intelligence (AI)-Based Metabolomics Data Processing and Analysis 27310.5 Future Direction of Metabolomics and Its Analysis 274References 27811 Strain Engineering, Flux Design, and Metabolic Production Using Big Data: Ongoing Advances and Opportunities 285Rafael S. Costa and Rui Henriques11.1 Introduction 28511.2 Big Data in Biotechnology: Prerequisites for ML-Based Approaches 28711.2.1 Data Multimodality and Heterogeneity 28711.2.2 Stratification and Data Transformations 28811.2.3 Handling of Missing Values and Outliers 28911.2.4 Longitudinal Studies 29011.3 Types of ML-Based Approaches for the Design of Microbial Cell Factories 29111.3.1 Machine Learning (ML) Models 29111.3.2 Supervised ml 29211.3.2.1 Neural Networks 29211.3.2.2 Decision Trees and Ensembles 29311.3.2.3 Alternative Predictive Approaches 29411.3.2.4 Selected Case Studies 29411.3.3 Unsupervised ml 29611.3.3.1 Clustering 29611.3.3.2 Biclustering 29711.3.3.3 Representation Learning and Dimensionality Reduction 29711.3.4 Hybrid ML and Constraint-Based Models 29811.3.4.1 CBM-FBA as Input for ml 29911.3.4.2 ml as Input for CBM-FBA 29911.4 ML-Based Approaches in Microbial Cell Factory Design: Case Studies 30011.4.1 ML-Based Approaches in Strain Engineering and Flux Design 30111.4.2 Application of ML-Based Approaches for Metabolic Production 30411.5 Conclusions and Perspectives 306References 30812 Next-Generation Metabolic Flux Analysis Using Machine Learning 317Ahmed Almunaifi, Richard C. Law, Samantha O’Keeffe, Kartikeya Pande, Tongjun Xiang, Onyedika Ukwueze, Aranaa Odai-Okley, Pin-Kuang Lai, and Junyoung O. Park12.1 Introduction 31712.2 Dynamic Nature of Metabolism 31712.3 Flux Balance Analysis and Metabolic Flux Analysis 32112.4 Incorporating Machine Learning into Metabolic Flux Analysis 32412.4.1 Challenges in Applying ML to MFA 32512.4.1.1 A Lack of Isotope Labeling Patterns and Flux Data for Training Machine Learning Models 32512.4.1.2 Variable-Size Input of Isotope Labeling Patterns for Flux Prediction 32612.4.1.3 Incorporation of Disparate Data Types 32612.4.1.4 Initial Computational Cost 32612.4.2 Available Computational Tools 32612.4.2.1 Sampling Metabolic Fluxes for Training ML Models 32712.4.2.2 Atom Mapping Throughout Metabolic Networks 32712.4.2.3 Selection of Information-Rich Isotope Tracers 32712.4.2.4 Other Helpful ML Tools 32812.4.3 A General Workflow for Machine Learning-Based Metabolic Flux Analysis 32812.4.3.1 Metabolic Model Construction 32912.4.3.2 Acquiring Training Data 33012.4.3.3 Training Machine Learning Models 33012.4.3.4 Evaluation of Machine Learning Models 33012.4.3.5 Execution: Computing Metabolic Fluxes Directly from Isotope Labeling Patterns 33112.4.4 Improved Speed and Accuracy of ML-Based MFA 33112.4.5 Toward Dynamic Flux and Isotope Labeling Analysis 33212.5 Future Outlook 333References 33413 Streamlining the Design-Build-Test-Learn Process in Automated Biofoundries 341Enrico Orsi, Nicolás Gurdo, and Pablo I. Nikel13.1 Introduction 34113.2 The Design-Build-Test-Learn Cycle (DBTLc) 34213.2.1 The DBTLc Components 34213.2.2 Application of the DBTLc for Strain Engineering 34513.2.3 Description of Biofoundries and Their Operating Parts 34513.2.4 Laboratory Workflows with Potential of Automation in Biofoundries 34713.3 Geographical Distribution of Biofoundries Around the Globe 35013.4 Challenges for Implementing Fully Automated Biofoundries 35213.5 Perspectives and Outlook 355Acknowledgments 357Ethics Declarations 357References 35714 Machine Learning-Enhanced Hybrid Modeling for Phenotype Prediction and Bioreactor Optimization 367Oliver Pennington, Yirong Chen, Youping Xie, and Dongda Zhang14.1 Bioprocess Modeling and Optimization 36714.1.1 Challenges in Bioprocess Modeling 36714.1.2 Machine Learning for Bioprocess Modeling 36914.1.2.1 Principal Component Analysis and Partial Least Squares 36914.1.2.2 Artificial Neural Networks 37014.1.2.3 Gaussian Processes 37014.1.2.4 Ensemble Learning 37114.1.2.5 Reinforcement Learning 37114.1.2.6 Future Prospects for Machine Learning 37114.1.3 Introduction to Hybrid Modeling 37214.1.3.1 Literature Review 37214.1.3.2 Hybrid Modeling for Biosystem Optimization 37314.2 Methodology for Hybrid Modeling 37814.2.1 Mechanistic Model Construction 37814.2.2 Time-Varying Parameter Estimation 38014.2.3 Machine Learning Model Construction 38114.2.4 Hybrid Model Uncertainty Estimation 38314.2.5 Considerations for Hybrid Model Construction 38414.2.5.1 Hybrid Model Greyness 38414.2.5.2 Discrepancy Hybrid Modeling 38414.2.5.3 Machine Learning Component 38514.2.6 Dynamic Optimization Using Hybrid Modeling 38514.2.6.1 Optimal Feeding in Fed-Batch Bioprocesses 38614.2.6.2 Dynamic Optimization Under Uncertainty 38614.3 Hybrid Modeling of Microalgal Lutein Production 38714.3.1 Introduction to Microalgal Lutein Production 38714.3.2 Experimental Setup and Data Availability 38814.3.3 Constructing a Hybrid Model for Microalgal Lutein Production 38914.3.3.1 Preliminary Kinetic Modeling of Microalgal Lutein Production 38914.3.3.2 Artificial Neural Network Implementation for Time-Varying Parameters 39114.3.4 Dynamic Optimization of Microalgal Lutein Production 39414.4 Summary 397References 398Index 407