Deep Learning In Biology And Medicine

Contents
Preface
About the Editors
Acknowledgements
1. Introduction
	1. Deep Learning
	2. Deep Learning in Biology and Medicine
	3. Book Outline
	References
2. Deep Learning for Medical Imaging
	1. Introduction
	2. Taxonomy of Deep Learning Strategies for Medical Image Analysis
		2.1. Input dimensionality
		2.2. Output dimensionality
		2.3. Input sources and modalities
		2.4. Network interconnection
		2.5. Contextual information
		2.6. Training schemes
	3. Applications in Brain Image Analysis
		3.1. Preprocessing
		3.2. Processing tasks
		3.3. Postprocessing
	4. Challenges and Future Directions
		4.1. Imaging challenges
		4.2. Deep learning challenges
	References
3. The Evolution of Mining Electronic Health Records in the Era of Deep Learning
	1. Introduction
	2. Data Organization and Structure of EHRs
		2.1. Brief history
		2.2. Benefits
		2.3. Data types and organization
		2.4. Data standardization and interoperability
	3. A Brief Introduction to Deep Learning
	4. Deep Learning with EHRs
		4.1. Disease prediction
		4.2. Disease phenotyping
		4.3. Patient stratification
		4.4. Clinical note understanding
	5. Discussion
	Acknowledgments
	References
4. Natural Language Technologies in the Biomedical Domain
	1. Introduction to Natural Language Processing
		1.1. NLP tasks
		1.2. NLP applications
	2. Empirical Methods in NLP
		2.1. Resources
	3. Statistical Methods in NLP
	4. NLP in the Biomedical Domain
		4.1. Characteristics of biomedical language
		4.2. Resources available in the biomedical domain
		4.3. Approaches for facing NLP tasks in the biomedical domain
		4.4. Linguistic preprocessing for NN models in the biomedical domain
	5. Neural Models for NLP: General Issues
		5.1. Embeddings
	6. Deep Learning Models for NLP
		6.1. Transformers
		6.2. Models
		6.3. Deep learning models for NLP in the biomedical domain
		6.4. Some examples of application of deep learning models in the biomedical domain
	7. Conclusions
	References
5. Metabolically Driven Latent Space Learning for Gene Expression Data
	1. Introduction
	2. Methods
		2.1. Genome-scale metabolic models
		2.2. Approximating FBA
		2.3. Dataset
			2.3.1. Real dataset
			2.3.2. Generating GE data
		2.4. VAEs
	3. Experimental Results
		3.1. FBA approximation
		3.2. Evaluating the VAE
		3.3. Is GEESE helping in reconstructing the latent space?
	4. Discussion and Conclusion
	References
6. Deep Learning in Cheminformatics
	1. Introduction
		1.1. QSAR/QSPR analysis
		1.2. De novo drug design
		1.3. Overview of this chapter
	2. Molecules and Their Representation
		2.1. Definitions
		2.2. Chemical and structural formulae of molecules
		2.3. Molecular descriptors
		2.4. Molecular fingerprints
		2.5. Other structural representations
	3. Deep Learning for QSAR/QSPR
		3.1. Problem
		3.2. Descriptor-based approaches
		3.3. Recursive neural network-based approaches
			3.3.1. Limitations
		3.4. Deep graph network-based approaches
	4. Deep Learning for De Novo Drug Design
		4.1. Problem
		4.2. Architectures
			4.2.1. Recurrent neural networks
			4.2.2. Variational autoencoders
			4.2.3. Generative adversarial networks
			4.2.4. Adversarial autoencoders
		4.3. Generation
			4.3.1. SMILES-based generation
			4.3.2. Matrix-based generation of molecular graphs
			4.3.3. Incremental generation of molecular graphs
			4.3.4. Discussion
			4.3.5. Sampling categorical distributions
			4.3.6. Evaluation metrics
		4.4. Optimization
			4.4.1. Conditional generation
			4.4.2. Latent space optimization
			4.4.3. Reinforcement learning
			4.4.4. Transfer learning
			4.4.5. Evaluation metrics
	5. Conclusions
	References
7. Deep Learning Methods for Network Biology
	1. Introduction
	2. Background Knowledge
		2.1. Network background and formalisation
		2.2. Learning problems on networks
		2.3. Ground concepts of systems biology
	3. Biological Networks and Publicly Available Resources
		3.1. Biological networks
			3.1.1. Protein–protein interaction network
			3.1.2. Drug–target network
			3.1.3. Gene expression network
			3.1.4. Gene regulatory network
			3.1.5. Brainnetwork
		3.2. Publicly available resources
	4. Deep Learning for Interactome (I)
		4.1. PPI prediction (PPIP)
		4.2. Essential protein prediction (EPP)
		4.3. Protein function prediction (PFP)
		4.4. Gene–disease association prediction (GDAP)
	5. Deep Learning for Network Pharmacology (NP)
		5.1. Drug–target interaction prediction (DTIP)
		5.2. Drug–disease association prediction (DDAP)
		5.3. Drug–drug interaction prediction (DDIP)
	6. Deep Learning for Other Biological Problems (BIO)
		6.1. miRNA–disease association prediction (MDAP)
		6.2. Disease analysis (DA)
		6.3. Brain analysis (BA)
	7. Conclusion and Future Work
	References
8. The Need for Interpretable and Explainable Deep Learning in Medicine and Healthcare
	1. Three Questions: From Data to Deep Learning
	2. Interpretability and Explainability in the Medical Context
	3. Making Deep Learning Interpretable in the Medical Domain: Approa
	4. Conclusions
	References
9. Ethical, Societal and Legal Issues in Deep Learning for Healthcare
	1. Introduction
		1.1. On the importance of AI ethics
	2. AI Ethical and Legal Guidelines Around the World
		2.1. US
		2.2. China
		2.3. EU
	3. EU’s Seven Requirements for Trustworthy AI
		3.1. Human agency and oversight
		3.2. Technical robustness and safety
		3.3. Privacy and data governance
		3.4. Transparency
		3.5. Diversity, non-discrimination and fairness
		3.6. Societal and environmental well-being
		3.7. Accountability
	4. The AI Application Lifecycle Stages
		4.1. Design stage
		4.2. Development stage
		4.3. Deployment and maintenance stage
		4.4. Usage stage
	5. Relevant EU Legislation
		5.1. Medical devices in EU
		5.2. Medical device malfunction
		5.3. Handling health data under the GDPR
			5.3.1. Further processing
	6. Technical Focus on Bias, Fairness, Explainability and Privacy in Deep Learning
		6.1. Biases in the data
		6.2. Fairness
		6.3. Interpretable and explainable AI
		6.4. Privacy
	7. Concluding Remarks
	References
Index