An AI platform to accelerate drug discovery: integrating target discovery, molecular generation, and simulation

Abstract

Declining research and development productivity is a structural challenge in the pharmaceutical industry, where the discovery, optimization, and clinical evaluation of a single medicine require long timelines, high cost, and repeated decision-making under uncertainty. Artificial intelligence (AI), big data analysis, and computational simulation are expected to accelerate drug discovery, but isolated applications cannot by themselves transform the whole process. This invited narrative review and platform perspective summarizes the concept and current implementation of an integrated Drug Discovery DX Platform (DXPF) that connects disease information and patient omics data to target gene/protein discovery, lead compound generation, and simulation-based evaluation. In the target-discovery component, Bayesian network and graph-based methods infer diseasespecific regulatory networks from omics data and support identification of mechanisms and candidate targets. In the drug-design component, graph convolutional neural network-based compound profile prediction is coupled with ChemTS molecular generation, docking, and molecular dynamics simulation to address multi-objective optimization of potency, absorption, distribution, metabolism, excretion (ADME), toxicity, and structural binding behavior. The platform also incorporates integrated databases and federated learning developed through academia-industry collaboration, enabling model improvement while preserving confidential company data. By connecting these modules through a browserbased interface, DXPF aims to shorten an end-to-end computational cycle from analysisready patient data to prioritized candidate structures; the approximately one-week objective is an engineering target under favorable input and computing conditions and excludes compound synthesis and biological, pharmacokinetic, toxicological, and clinical validation. Current evidence is mainly module-level, retrospective, or computational, and prospective benchmarking, uncertainty assessment, and experimental validation are required before effects on attrition or regulatory decision-making can be established. Future expansion to biologics, disease-prevention AI, and preclinical and clinical data could broaden the platform's scope, but these extensions remain development goals.