Deep Learning in Bioinformatics: Techniques and Applications in Practice

Front Cover
Deep Learning in Bioinformatics
Copyright
Contents
Acknowledgments
Preface
1 Why life science?
1.1 Introduction
1.2 Why deep learning?
1.3 Contemporary life science is about data
1.4 Deep learning and bioinformatics
1.5 What will you learn?
2 A review of machine learning
2.1 Introduction
2.2 What is machine learning?
2.3 Challenge with machine learning
2.4 Overfitting and underfitting
2.4.1 Mitigating overfitting
2.4.2 Adjusting parameters using cross-validation
2.4.3 Cross-validation methods
2.5 Types of machine learning
2.5.1 Supervised learning
2.5.2 Unsupervised learning
2.5.3 Reinforcement learning
2.6 The math behind deep learning
2.6.1 Tensors
2.6.2 Relevant mathematical operations
2.6.3 The math behind machine learning: statistics
2.7 TensorFlow and Keras
2.8 Real-world tensors
2.9 Summary
3 An introduction of Python ecosystem for deep learning
3.1 Basic setup
3.2 SciPy (scientific Python) ecosystem
3.3 Scikit-learn
3.4 A quick refresher in Python
3.4.1 Identifier
3.4.2 Comments
3.4.3 Data type
3.4.4 Control flow statements
3.4.5 Data structure
3.4.6 Functions
3.5 NumPy
3.6 Matplotlib crash course
3.7 Pandas
3.8 How to load dataset
3.8.1 Considerations when loading CSV data
3.8.2 Pima Indians diabetes dataset
3.8.3 Loading CSV files in NumPy
3.8.4 Loading CSV files in Pandas
3.9 Dimensions of your data
3.10 Correlations between features
3.11 Techniques to understand each feature in the dataset
3.11.1 Histograms
3.11.2 Box-and-whisker plots
3.11.3 Correlation matrix plot
3.12 Prepare your data for deep learning
3.12.1 Scaling features to a range
3.12.2 Data normalizing
3.12.3 Binarize data (make binary)
3.13 Feature selection for machine learning
3.13.1 Univariate selection
3.13.2 Recursive feature elimination
3.13.3 Principal component analysis
3.13.4 Feature importance
3.14 Split dataset into training and testing sets
3.15 Summary
4 Basic structure of neural networks
4.1 Introduction
4.2 The neuron
4.3 Layers of neural networks
4.4 How a neural network is trained?
4.5 Delta learning rule
4.6 Generalized delta rule
4.7 Gradient descent
4.7.1 Stochastic gradient descent
4.7.2 Batch gradient descent
4.7.3 Mini-batch gradient descent
4.8 Example: delta rule
4.8.1 Implementation of the SGD method
4.8.2 Implementation of the batch method
4.9 Limitations of single-layer neural networks
4.10 Summary
5 Training multilayer neural networks
5.1 Introduction
5.2 Backpropagation algorithm
5.3 Momentum
5.4 Neural network models in keras
5.5 `Hello world!' of deep learning
5.6 Tuning hyperparameters
5.7 Data preprocessing
5.7.1 Vectorization
5.7.2 Value normalization
5.8 Summary
6 Classification in bioinformatics
6.1 Introduction
6.1.1 Binary classification
6.1.2 Pima indians onset of diabetes dataset
6.1.2.1 Import libraries
6.1.2.2 Load data
6.1.2.3 Keras model
6.1.2.4 Compile the model
6.1.2.5 Fit the model
6.1.2.6 Evaluate the model
6.1.2.7 Tie it all together
6.1.2.8 Make predictions
6.1.3 Label encoding
6.2 Multiclass classification
6.2.1 Sigmoid and softmax activation functions
6.2.2 Types of classification
6.3 Summary
7 Introduction to deep learning
7.1 Introduction
7.2 Improving the performance of deep neural networks
7.2.1 Vanishing gradient
7.2.2 Overfitting
7.2.2.1 Reducing the network's size
7.2.2.2 Dropout
7.2.2.3 Weight regularization
7.2.3 Computational load
7.3 Configuring the learning rate in keras
7.3.1 Adaptive learning rate
7.3.2 Layer weight initializers
7.4 Imbalanced dataset
7.5 Breast cancer detection
7.5.1 Goals
7.5.2 Introduction and task definition
7.5.3 Implementation
7.5.3.1 Loading, preprocessing, preparations for modeling
7.5.3.2 Fully connected neural network (FCNN)
7.5.3.3 Adding dropout to the network (FCNN + dropout)
7.5.3.4 Adding L2 weight regularization (FCNN + L2)
7.5.3.5 Adding L2 weight regularization and dropout (FCNN + L2 + dropout)
7.5.3.6 Adding L1_L2 weight regularization (FCNN + L1_L2)
7.5.3.7 Reducing the size of the network
7.5.3.8 Summary
7.6 Molecular classification of cancer by gene expression
7.6.1 Goals
7.6.2 Introduction and task definition
7.6.3 Implementation
7.6.3.1 Loading, preprocessing, preparations for modeling
7.6.3.2 Dimension reduction using principal component analysis (PCA)
7.6.3.3 Model
7.7 Summary
8 Medical image processing: an insight to convolutional neural networks
8.1 Convolutional neural network architecture
8.2 Convolution layer
8.3 Pooling layer
8.4 Stride and padding
8.5 Convolutional layer in keras
8.6 Coronavirus (COVID-19) disease diagnosis
8.6.1 Goals
8.6.2 Introduction and task definition
8.6.3 Implementation
8.6.3.1 Importing required libraries
8.6.3.2 Plotting some instances of the dataset
8.6.3.3 Defining the model
8.6.3.4 Discussing the relevance of deep learning for small-data problems
8.6.3.5 Predicting covid-19
8.6.4 Conclusion
8.7 Predicting breast cancer
8.7.1 Goals
8.7.2 Introduction and task definition
8.7.3 Implementation
8.7.3.1 Importing required libraries
8.7.3.2 Looking for all available directories in Kaggle account
8.7.3.3 Plotting images using cv2 module
8.7.3.4 Finding specific pattern in the name of images
8.7.3.5 Preprocessing data
8.7.3.6 Dealing with imbalanced data
8.7.3.7 Defining the sequential model
8.7.4 Conclusion
8.8 Diabetic retinopathy detection
8.8.1 Goals
8.8.2 Introduction and task definition
8.8.3 Implementation
8.8.3.1 Importing required libraries and reading the data
8.8.3.2 Preprocessing data
8.8.3.3 Defining model based on functional API
8.8.3.4 Defining another model using ResNet50 model
8.8.4 Conclusion
8.9 Summary
9 Popular deep learning image classifiers
9.1 Introduction
9.2 LeNet-5
9.3 AlexNet
9.4 ZFNet
9.5 VGGNet
9.6 GoogLeNet/inception
9.7 ResNet
9.8 DenseNet
9.9 SE-Net
9.10 Summary
10 Electrocardiogram (ECG) arrhythmia classification
10.1 Introduction
10.2 MIT-BIH arrhythmia database
10.3 Preprocessing
10.4 Data augmentation
10.5 Architecture of the CNN model
10.6 Summary
11 Autoencoders and deep generative models in bioinformatics
11.1 Introduction
11.2 Autoencoders
11.2.1 Encoder
11.2.2 Decoder
11.2.3 Distance function
11.3 Variant types of autoencoders
11.3.1 Undercomplete autoencoders
11.3.2 Deep autoencoders
11.3.3 Convolutional autoencoders
11.3.4 Sparse autoencoders
11.3.5 Denoising autoencoders
11.3.6 Variational autoencoders
Intuition
VAE is a generative model
How does a variational autoencoder work?
Creating decoder
Building the architecture of the VAE: connecting the encoder and decoder
Defining loss function and compiling model
11.3.7 Contractive autoencoders
11.4 An example of denoising autoencoders – bone suppression in chest radiographs
11.4.1 Architecture
11.5 Implementation of autoencoders for chest X-ray images (pneumonia)
11.5.1 Undercompleted autoencoder
11.5.2 Sparse autoencoder
11.5.3 Denoising autoencoder
11.5.4 Variational autoencoder
11.5.5 Contractive autoencoder
11.6 Generative adversarial network
11.6.1 GAN network architecture
11.6.2 GAN network cost function
11.6.3 Cost function optimization process in GAN
11.6.4 General GAN training process
11.7 Convolutional generative adversarial network
11.7.1 Deconvolution layer
11.7.2 DCGAN network structure
11.8 Summary
12 Recurrent neural networks: generating new molecules and proteins sequence classification
12.1 Introduction
12.2 Types of recurrent neural network
12.3 The problem, short-term memory
12.4 Bidirectional LSTM
12.5 Generating new molecules
12.5.1 Simplified molecular-input line-entry system
12.5.2 A generative model for molecules
12.5.3 Generating new SMILES
12.5.4 Analyzing the generative model's output
12.6 Protein sequence classification
12.6.1 Protein structure
12.6.2 Protein function
12.6.3 Prediction of protein function
12.6.4 LSTM with dropout
12.6.5 LSTM with bidirectional and CNN
12.7 Summary
13 Application, challenge, and suggestion
13.1 Introduction
13.2 Legendary deep learning architectures, CNN, and RNN
13.3 Deep learning applications in bioinformatics
13.4 Biological networks
13.4.1 Learning tasks on graphs
13.4.2 Graph neural networks
13.5 Perspectives, limitations, and suggestions
13.6 DeepChem, a powerful library for bioinformatics
13.7 Summary
Index
Back Cover