Machine/Deep Learning Causal Pharmaco-Analytics for Preclinical System Pharmacology Modeling and Clinical Outcomes Analytics

The modern drug discovery and development pipeline are complex, lengthy, and costly. The amount and availability of biomedical data explosive increase in the past decade, accompanied by rapid developments in computational technology. Those provide new and exciting opportunities to better and systematically understand the biology and pharmacology of diseases with the help of data science, machine/deep learning (ML/DL), and artificial intelligence (AI) technologies. Pharmaco-Analytics rises from introducing data science-driven methods to the traditional modeling and simulation in pharmaceutical & clinical sciences. It encompasses topics that cover preclinical and clinical analyses, for example, computational drug discovery, bioanalytical methodology, Pharmacometrics and Systems Pharmacology, Pharmacoeconomics, and outcomes analytics. In the Pharmaco-Analytics field, there are emerging AI/ML technology development and growing numbers of applications published increasingly facilitate pharmaceutical sciences and health care research. We present six studies in Chapters 2 to 4 introducing our innovation of developing ML/AI methods to inform preclinical modeling and clinical outcomes research. The first two studies describe two computational methods on target identification using Pharmaco-Analytics technology. The first study involves the development of an AI platform to investigate drug abuse Poly-pharmacology using computational chemistry and machine learning algorithms. The second study introduces a novel algorithm (DeepTargetHunter) to identify the target of small molecules based on a novel deep learning technique for drug repurposing. The subsequent two studies focus on developments for preclinical properties prediction. The first study introduces a novel graph-based method (DeepGhERG), to predict the hERG cardiotoxicity of small molecules and the second study describes DL methods to predict blood-brain barrier permeability. Lastly, we examine two methods to predict the risk of substance use disorder (SUD) based on childhood psychopathological traits. The first study presents a novel ML method to predict SUD outcomes based on deriving 30 of the most important questionnaire items predicting SUD. Whereas the second study introduces a novel approach called CausalSUD to identify the causal relationship between psychopathological cluster patterns and risk of SUD from late childhood to adulthood. In aggregate, the results from this research demonstrate the heuristic utility of AI/ML methods for advancing the Pharmaco-Analytics research in preclinical modeling and causal machine learning on clinical outcomes analysis