AI-Enabled Drug Discovery in Pharmaceutical Sciences: A Systematic Review of Molecular Models, Target Prediction, Lead Optimization, and Translational Barriers

Download PDF

Artificial intelligence has become a major methodological force in pharmaceutical drug discovery, supporting molecular modeling, compound prioritization, target prediction, virtual screening, de novo design, ADMET estimation, and lead optimization. This systematic review evaluates how AI-enabled methods are used across drug discovery tasks and how far their outputs can be interpreted as translationally meaningful evidence. The review was designed around focused questions concerning molecular representations, model families, target and drug–target prediction, lead optimization workflows, validation strategies, and barriers to pharmaceutical implementation. The reviewed evidence covers machine learning, deep learning, graph neural networks, transformer architectures, generative models, multitask learning, transfer learning, explainable AI, benchmark resources, and structure-based computational approaches. Particular attention is given to the distinction between retrospective benchmark performance, in silico prioritization, prospective experimental validation, preclinical relevance, clinical translation, and regulatory readiness. The synthesis indicates that AI can strengthen hypothesis generation, expand searchable chemical space, prioritize compounds and targets, and improve decision support in early discovery. However, reliable translation depends on dataset quality, leakage-aware benchmarking, assay context, biological plausibility, reproducibility, interpretability, uncertainty characterization, and experimental confirmation. The review contributes a systematic evidence framework linking molecular model design to validation needs and translational barriers. Future pharmaceutical AI research should move beyond isolated benchmark improvement toward robust, explainable, externally validated, and experimentally integrated discovery workflows.