Artificial Neural Networks

Preface
Contents
Contributors
Chapter 1: Identifying Genotype-Phenotype Correlations via Integrative Mutation Analysis
	1 Introduction
	2 Materials
		2.1 Data Curation
			2.1.1 Mutation Curation
			2.1.2 Protein Structure Curation
		2.2 An Overview of Computational Tools to Analyze Missense Mutations
	3 Methods
		3.1 Predicting and Analyzing Structural and Biophysical Effects of Mutations Using the mCSM Platform
		3.2 mCSM Platform Output
			3.2.1 Arpeggio
			3.2.2 MTR-Viewer
				Gene Viewer
				Variant Query
			3.2.3 mCSM-Stability/PPI/DNA
				Single Mutation
				Multiple or Systematic
			3.2.4 SDM
				Single Mutation
				Multiple Mutations
			3.2.5 DUET
				Single Mutation
				Systematic Mutations
			3.2.6 DynaMut
				Single Mutation
				Multiple Mutations
			3.2.7 mCSM-PPI2
				Single Mutation
				List Mutation
				Alanine Scanning
				Saturation Mutagenesis
			3.2.8 mCSM-NA
				Single Prediction
				List Mutation
			3.2.9 mCSM-lig
		3.3 Identification of Driving Molecular Consequences
		3.4 Machine Learning Phenotypes: Building a Predictive Classifier
	4 Notes
	References
Chapter 2: Machine Learning for Biomedical Time Series Classification: From Shapelets to Deep Learning
	1 Introduction
		1.1 Challenges of Biomedical Time Series Data Analysis
			1.1.1 Time Series Signals
			1.1.2 Time Series Labels
			1.1.3 Irregular Sampling
		1.2 Notation
		1.3 Our Running Example: Sepsis Prediction
	2 Time Series Classification Tasks
		2.1 Whole Time Series Classification
		2.2 Onset Detection
			2.2.1 TimePointWise Classification
			2.2.2 Time Series Classification on Time Series Windows
	3 Methods for Time Series Classification
		3.1 Distance-Based Time Series Classification
			3.1.1 Dynamic Time Warping and Nearest Neighbors
				k-nearest Neighbor Algorithm
			3.1.2 Shapelets
				Collect Shapelet Candidates
				Compute Distances Between Candidates and Time Series
				Find the Best Candidate
					Significant Subsequence Mining
				Collect Shapelet Candidates
				Assess the Statistical Significance of Shapelet Candidates
				Multiple Hypothesis Correction
		3.2 Deep Learning for Time Series Classification
			3.2.1 Multilayer Perceptron (MLP)
			3.2.2 Recurrent Neural Networks (RNNs)
				Long Short-Term Memory
			3.2.3 Residual Networks (ResNets)
			3.2.4 Convolutional Neural Networks (CNNs)
			3.2.5 Temporal Convolutional Networks (TCNs)
		3.3 Tools
	4 Summary
	References
Chapter 3: Siamese Neural Networks: An Overview
	1 Introduction
	2 Siamese Neural Network: The Architecture
	3 Applications and Theoretical Studies
		3.1 Audio and Speech Signal Processing
		3.2 Biology
		3.3 Chemistry and Pharmacology
		3.4 Geometry and Graphics
		3.5 Image Analysis
			3.5.1 Handwritten Forgery Detection
			3.5.2 One-Shot Image Recognition
		3.6 Medicine and Health
		3.7 Optics and Physics
		3.8 Robotics
		3.9 Sensor-Based Activity Recognition
		3.10 Software Development
		3.11 Text Mining and Natural Language Processing (NLP)
		3.12 Video Analysis
		3.13 Theory Studies
	4 Software Packages, Tutorials, and Guides
		4.1 Software Packages
		4.2 Tutorials
		4.3 Guides
	5 Discussion and Conclusions
	References
Chapter 4: Computational Methods for Elucidating Gene Expression Regulation in Bacteria
	1 Introduction
	2 Materials
		2.1 Hardware
		2.2 sRNA Detection and Prioritization
			2.2.1 Data
			2.2.2 Software
		2.3 sRNA Target Prediction
			2.3.1 Data
			2.3.2 Software
		2.4 Promoter Recognition
			2.4.1 Data
			2.4.2 Software
	3 Methods
		3.1 sRNA Detection and Prioritization
		3.2 sRNA Target Prediction
			3.2.1 CopraRNA
			3.2.2 SPOT (See Note 7)
		3.3 Promoter Recognition
			3.3.1 bTSSfinder
			3.3.2 G4PromFinder
	4 Notes
	References
Chapter 5: Neuroevolutive Algorithms Applied for Modeling Some Biochemical Separation Processes
	1 Modeling Methodology Based on Neural Networks and Differential Evolution Algorithm
		1.1 Artificial Neural Networks
		1.2 Differential Evolution Algorithm
		1.3 Development of an Optimal Neural Network with  DE
	2 Differential Evolution-Based Neuroevolutive Applications
	3 Practical Examples
		3.1 Reactive Extraction of Folic Acid (Vitamin B9)
		3.2 Facilitated Pertraction of Vitamin  C
	4 Notes
	References
Chapter 6: Computational Approaches for De Novo Drug Design: Past, Present, and Future
	1 Introduction
	2 De Novo Drug Design
		2.1 Molecular Representations
		2.2 Multiple Objectives
	3 Optimization Methods
		3.1 Evolutionary Algorithms
		3.2 Particle Swarm Optimization (PSO)
		3.3 Simulated Annealing
	4 Deep Learning Algorithms
		4.1 Recurrent Neural Networks (RNNs)
		4.2 Variational Autoencoders
		4.3 Deep Reinforcement Learning
		4.4 Generative Adversarial Networks
	5 Competition or Cooperation?
	6 Conclusion and Perspective
	References
Chapter 7: Data Integration Using Advances in Machine Learning in Drug Discovery and Molecular Biology
	1 Introduction
	2 Deep Learning: Next-Generation Machine Learning
		2.1 Deep Neural Networks
		2.2 Drug Discovery
		2.3 Application of Convolutional Neural Networks to Predict Ligand-Protein Interactions
		2.4 Application of Deep Learning in Compound Property and Activity Prediction
		2.5 De Novo Design Through Deep Learning
		2.6 Network Biology
		2.7 Network Embedding
	3 Challenges for the Future in Network Biology
		3.1 Dynamic Networks
		3.2 Hierarchical Network Structure
		3.3 Heterogeneous Networks
		3.4 Scalability
		3.5 Interpretability
	4 Conclusion
	References
Chapter 8: Building and Interpreting Artificial Neural Network Models for Biological Systems
	1 Introduction
	2 Methods
		2.1 Modeling a Biological System Using a Neural Network
		2.2 Feature Selection
		2.3 Model Building
		2.4 Evaluation of Neural Network Models
		2.5 Calliper Randomization for Interpreting Neural Network Models
		Algorithm 1 Calliper Randomization Algorithm
	3 Summary
	References
Chapter 9: A Novel Computational Approach for Biomarker Detection for Gene Expression-Based Computer-Aided Diagnostic Systems ...
	1 Introduction
	2 Materials and Methods
		2.1 Bi-biological Filter: Preliminary Filtering
			2.1.1 Neither cancer nor healthy biomarker filtering
				Module Extraction
				Selection of Shared Clusters
			2.1.2 Removal of Healthy Biomarkers
		2.2 Best First Search and SVM with Fivefold Cross-Validation Wrapper
		2.3 Classification and Evaluation
	3 Results and Discussion
		3.1 Feature Selection
			3.1.1 Neither Cancer nor Healthy Biomarker Filter
			3.1.2 Removal of Healthy Biomarkers
			3.1.3 Best First Search and SVM with Fivefold Cross-Validation Wrapper
		3.2 Classifications
	4 Conclusions
	References
Chapter 10: Applying Machine Learning for Integration of Multi-Modal Genomics Data and Imaging Data to Quantify Heterogeneity ...
	1 Introduction
	2 Background
		2.1 Neural Network for Genomics Data Analysis
		2.2 Convolutional Neural Network (CNN) for Pathological Image Analysis
		2.3 Deep Learning Methods for Combining Genomics and Imaging Data
			2.3.1 Transfer Learning
			2.3.2 Autoencoder
	3 Applying Deep Learning for Mining Biomedical Imaging and Molecular Genomics Data
		3.1 Interpretable Deep Learning in Medical Research
		3.2 Suitable Loss Functions for Biomedical Data
			3.2.1 Cross-Entropy
			3.2.2 Weighted Cross Entropy
			3.2.3 Focal Loss
			3.2.4 Dice Loss
			3.2.5 Combinations
	4 Improving Deep NN Implementation for Biomedical Genomics and Imaging Data
		4.1 Integrating Genomics Data and Tissue Imaging Data for Classifying Cancer Cell Types from H&E Images
		4.2 Applications of Spatial Omics and Deep Learning in Characterizing Cancer Heterogeneity
	5 Conclusions
	References
Chapter 11: Leverage Large-Scale Biological Networks to Decipher the Genetic Basis of Human Diseases Using Machine Learning
	1 Introduction
		1.1 Traditional Genetics Approaches to Identify Disease-Associated Genetic Variants
		1.2 Whole-Genome Annotation of Regulatory Elements
		1.3 Regulatory Networks Are Needed to Reveal the Mechanisms
		1.4 Machine Learning Developments to Integrate Regulatory Networks for Advanced  GWAS
			1.4.1 Predict Regulatory Elements and Prioritize Disease-Associated  SNPs
			1.4.2 Predict TF-Enhancer Links and Identify Master Regulators Disrupted by  SNPs
			1.4.3 Predict Enhancer-Gene Links and Discover Dysregulated Genes by  SNPs
	2 Software and Data Resources
	3 Methods
		3.1 Regulatory Element Prediction and Improved Fine-Mapping of Disease  SNPs
			3.1.1 ChromHMM Algorithm
			3.1.2 bfGWAS Algorithm
		3.2 TF Binding Imputation and Incorporation with Genetic Variants
			3.2.1 Catchitt Algorithm
			3.2.2 deltaSVM Algorithm
		3.3 Enhancer-Gene Link Prediction and Identify Disease-Associated Genes
			3.3.1 IM-PET Algorithm
			3.3.2 ARVIN Algorithm
		3.4 Examples Illustrating Results from the Pipeline
	4 Notes
	References
Chapter 12: Predicting Host Phenotype Based on Gut Microbiome Using a Convolutional Neural Network Approach
	1 Introduction
		1.1 PopPhy CNN Network
		1.2 PopPhy-CNN Framework
		1.3 Learning Model in PopPhy-CNN
		1.4 Feature Ranking in PopPhy-CNN
	2 Materials
		2.1 Collected Microbiome  Data
		2.2 Required Computational Tools for PopPhy-CNN
	3 Methods
		3.1 Preparing the Matrix Input
			3.1.1 Input Format
			3.1.2 Preprocessing Data
			3.1.3 Generating Matrix Representations of Populated Taxonomic  Tree
		3.2 Training the CNN Model
			3.2.1 Constructing the CNN Architecture
			3.2.2 Normalizing the  Data
			3.2.3 Training a PopPhy-CNN Model
		3.3 Evaluating and Visualizing Taxonomic Features
			3.3.1 Scoring Taxonomic Features
			3.3.2 Generating the Network  File
	4 A Demonstration Using a Cirrhosis Dataset
		4.1 Using PopPhy-CNN
		4.2 The Ranked List of Microbial  Taxa
		4.3 Visualizing Feature Scores
	5 Conclusions
	6 Notes
	References
Chapter 13: Predicting Hot Spots Using a Deep Neural Network Approach
	Abbreviations
	1 Introduction
	2 Materials
		2.1 Data
		2.2 Tools
	3 Methods
		3.1 Feature Extraction
		3.2 Deep Learning Classification
		3.3 Metrics Used for Evaluating Model Performance
	4 Notes
	References
Chapter 14: Using Neural Networks for Relation Extraction from Biomedical Literature
	1 Introduction
	2 Related  Work
		2.1 Natural Language Processing
		2.2 Text Mining Primary Tasks
		2.3 Initial Approaches for Relation Extraction
	3 Neural Networks for Relation Extraction
		3.1 Architectures
		3.2 Data Representations
		3.3 Ontologies
	4 Evaluation Measures
	5 Conclusions
	References
Chapter 15: A Hybrid Levenberg-Marquardt Algorithm on a Recursive Neural Network for Scoring Protein Models
	1 Introduction
	2 Material and Methods
	3 Results
	4 Discussion
	5 Conclusions
	References
Chapter 16: Secure and Scalable Collection of Biomedical Data for Machine Learning Applications
	1 Scalable Data Collection
		1.1 On-Premise Versus Cloud Deployments
		1.2 Operating System Constraints and Considerations
		1.3 Network Topology Considerations
		1.4 Virtualized Systems
		1.5 Instrument-Specific Data-Writing Behaviors
		1.6 Impact of Size and Number of Files on Data Collection Strategy
		1.7 Compression
		1.8 Bandwidth Considerations
		1.9 Substitute for Direct Instrument Connectivity
	2 Secure Data Collection
		2.1 Threat Model and Landscape
		2.2 Single-Tenant Versus Multitenant Considerations
		2.3 End-to-End Encryption
		2.4 Achieving Encryption in Transit
		2.5 Additional Data Encryption
		2.6 Achieving Encryption At Rest
		2.7 Encryption and Compression
	3 Metadata Extraction
		3.1 What Is ``Metadata´´?
		3.2 File Format Diversity in the Life Sciences
		3.3 Reverse Engineering File Formats Manually to Identify Metadata of Interest
		3.4 Storing Metadata
		3.5 Designing for Interoperability Using Application Programming Interfaces (APIs)
	References
Chapter 17: AI-Based Methods and Technologies to Develop Wearable Devices for Prosthetics and Predictions of Degenerative Dise...
	1 Introduction
	2 Bioinspired Softcomputing Methods
		2.1 Evolving Fuzzy Neural Networks
		2.2 Spiking Neural Networks
	3 Technologies
		3.1 Analog Front-End
		3.2 Sensors
			3.2.1 Photometric Sensors
		3.3 Inertial Sensors
		3.4 Application Specific Processors
		3.5 Biomedical Sensor Front-End
	4 Application Frameworks
		4.1 Prosthesis
		4.2 Prediction
	References
Index