The Era of Artificial Intelligence, Machine Learning, and Data Science in the Pharmaceutical Industry

Front Matter
Copyright
Contributors
Preface
Acknowledgments and conflicts of interest
Introduction to drug discovery
	The drug discovery process
		Target identification
		Target validation
		Hit identification and lead discovery
			Virtual screening
			Compound libraries
			High-throughput screening
			Structure-based drug discovery
			Fragment-based drug discovery
			Phenotypic drug discovery
			Natural products
		Lead optimization
			Modeling in lead optimization
		Precision medicine
		Clinical testing and beyond
	References
Introduction to artificial intelligence and machine learning
	Supervised learning
	Unsupervised learning
	Semisupervised learning
	Model selection
	Types of data
	Other key considerations
		Feature generation and selection
		Censored and missing data
		Dependencies in the data: Time series or sequences, spatial dependence
	Deep learning
	Uncertainty quantification
		Bayesian inference
	References
Data types and resources
	Notes on data
	Omics data
		Genomics
		Transcriptomics
		Metabolomics and lipomics
		Proteomics
	Chemical compounds
		SDF format
		InChI and InChI Key format
		SMILES and SMARTS format
		Fingerprint format
		Other descriptors
		Similarity measures
	QSAR with regards to safety
	Data resources
		Toxicity related databases
		Drug safety databases
		Key public data-resources for precision medicine
			Resources for enabling the development of computational models in oncology
			Key genomic/epigenomic resources for therapeutic areas other than oncology
			Resources for accessing metadata and analysis tools
	References
Target identification and validation
	Introduction
	Target identification predictions
	Gene prioritization methods
	Machine learning and knowledge graphs in drug discovery
		Introduction
		Graph theory algorithms
		Graph-oriented machine learning approaches
			Feature extraction from graph
			Graph-specific deep network architectures
		Drug discovery knowledge graph challenges
	Data, data mining, and natural language processing for information extraction
		What is natural language processing
		How is it used for drug discovery and development
		Where is it used in drug discovery and development (and thoughts on where it is going at the end)
	References
Hit discovery
	Chemical space
	Screening methods
	High-throughput screening
	Computer-aided drug discovery
		De novo design
		Virtual screening
		Data collection and curation
		Databases and access
		Compounds
		Targets
		Activity measurement
		Cleaning collected data—Best practices
	Representing compounds to machine learning algorithms
	Candidate learning algorithms
		Naive Bayes
		k-Nearest neighbors
		Support vector machines
		Random forests
		Artificial neural networks
		Multitask deep neural networks
	Future directions: Learned descriptors and proteochemometric models
		Graph convolutional and message passing neural networks
		Proteochemometric models
	Evaluating virtual screening models
		Train-test splits: Random, temporal, or cluster-based?
		External validation
		Prospective experimental validation
	Clustering in hit discovery
		Butina clustering
		K-means clustering
		Hierarchical clustering
	References
Lead optimization
	What is lead optimization
	Applications of machine learning in lead optimization
	Assessing ADMET and biological activities properties
	Matched molecular pairs
		Machine learning with matched molecular pairs
	References
Evaluating safety and toxicity
	Introduction to computational approaches for evaluating safety and toxicity
	In silico nonclinical drug safety
	Machine learning approaches to toxicity prediction
		k-nearest neighbors
		Logistic regression
		Svm
		Decision tree
		Random forest and other ensemble methods
		Naïve Bayes classifier
		Clustering and primary component analysis
		Deep learning
	Pharmacovigilance and drug safety
		Data sources
		Disproportionality analysis
		Mining medical records
		Electronic health records
		Social media signal detection
		Knowledge-based systems, association rules, and pattern recognition
	Conclusions
	References
Precision medicine
	Cancer-targeted therapy and precision oncology
	Personalized medicine and patient stratification
		Methods for survival analysis
	Finding the “right patient”: Data-driven identification of disease subtypes
		Subtypes are the currency of precision medicine
		The nature of clusters and clustering
		Selection and preparation of data
		Approaches to clustering and classification
			Unsupervised and supervised partitional classification
			Hierarchical classification
			Biclustering
			Clustering trajectories and time series
			Integrative analysis
			Deep approaches
		Validation and interpretation
			Direct validation
			Indirect validation
			Characterization
	Key advances in healthcare AI driving precision medicine
		Key challenges for AI in precision medicine
	References
Image analysis in drug discovery
	Cells
	Spheroids
	Microphysiological systems
	Ex vivo tissue culture
	Animal models
		Tissue pathology
	Aims and tasks in image analysis
		Image enhancement
		Image segmentation
	Region segmentation in digital pathology
		Why is it used?
			The reduction in time to build acute models compared with rule-based solutions is significant
			Reduction in pathologist and scientist time doing manual aspects of annotation and analysis
			Consistency of decision making (inter and intrauser error)
	Feature extraction
		Image classification
			Limitations and barriers to using DL in image analysis
	The status of imaging and artificial intelligence in human clinical trials for oncology drug development
		Computational pathology image analysis
		Radiology image analysis
			AI-based radiomics to predict response to therapy
				Protein kinase inhibitors
				Chemotherapy/chemoradiotherapy
				Immunotherapy
			Challenges in applying radiomics to drug discovery
				Clinical trial validation
				Regulatory approval
				Distribution and reimbursement
		Conclusion
	Future directions
		Imaging for drug screening
		Computational pathology and radiomics
	References
Clinical trials, real-world evidence, and digital medicine
	Introduction
		The importance of ethical AI
	Clinical trials
		Site selection
		Recruitment modeling for clinical trials
			Recruitment start dates
			The Poisson Gamma model of trial recruitment
			Nonhomogeneous recruitment rates
		Applications of recruitment modeling in the clinical supply chain
		Clinical event adjudication and classification
		Identifying predictors of treatment response using clinical trial data
	Real-world data: Challenges and applications in drug development
		The RWD landscape
		Barriers for adoption of RWD for clinical research
			Data quality
			Interoperability
		Use of RWE/RWD in clinical drug development and research
		Concluding thoughts on RWD
	Sensors and wearable devices
		Sample case study: Parkinson’s disease
		Standards and regulations and concluding thoughts
	Conclusions
	References
Beyond the patient: Advanced techniques to help predict the fate and effects of pharmaceuticals in the environment
	Overview
	Background
	Current European and US legislation for environmental assessment of pharmaceuticals
	Animal testing for protecting the environment
	Issues for database creation
	Opportunities to refine animal testing for protecting the environment
	Current approaches to predicting uptake of pharmaceuticals
	What makes pharmaceuticals special?
	Why do pharmaceuticals effect wildlife?
	What happens in the environment?
	Predicting uptake using ML
	Regional issues and the focus of concern
	Intelligent regulation—A future state of automated AI assessment of chemicals
	Key points for future development
	References
Index