AI and Real-World Evidence for Drug Discovery, Repurposing, and Safety

Finding new drugs, and finding new uses for existing drugs, remains slow, expensive, and associated with a high failure rate. Research now has access to more data than ever before, including electronic health records, claims databases, registries, multi-omics and functional genomics, and molecular and cellular assays. Modern AI and machine learning enable scalable pattern recognition and strong predictive performance across these modalities, and real-world evidence (RWE) derived from routine clinical data offers an increasingly central role in informing drug development decisions.Yet prediction and association alone are often insufficient for translation. Many questions in drug discovery and repurposing are fundamentally causal: whether modulating a target changes disease biology, whether a therapy improves patient outcomes in routine care, which populations benefit or experience harm, and how drug combinations modify safety and effectiveness. These questions are difficult because observational clinical data are shaped by confounding, selection bias, and evolving care processes, and findings from experimental systems may not generalise to human populations without explicit assumptions.The central problem addressed in this Research Topic is the gap between AI-generated signals and the robust evidence required for reliable drug discovery, repurposing, and safety decisions. Many candidates fail because observed associations do not translate into true clinical benefit, because safety risks emerge under real-world polypharmacy, or because findings from experimental systems do not generalise to human populations. This Research Topic will highlight methods and applications that bring together AI, machine learning, and real-world evidence — including causal inference, deep learning, and representation learning — to generate evidence that is credible and decision-relevant.Contributions should be grounded in clearly defined research questions, appropriate data sources, and transparent analytic assumptions. Purely in silico or computational studies are expected to include experimental, clinical, or real-world validation, or to provide a clear and explicit pathway demonstrating how predictions connect to testable biological or clinical hypotheses. Emphasis is placed on rigorous evaluation, including robustness checks, sensitivity analysis, uncertainty quantification, and assessment of generalisability across settings and populations. The Topic also promotes reproducibility through clear reporting of data provenance and limitations, and, where feasible, sharing of code and usable research artefacts.Articles focusing on the following themes are particularly welcomed:AI and machine learning for drug discovery and repurposing, including deep learning, graph neural networks, large language models, and foundation models applied to biomedical dataReal-world evidence for drug effectiveness and safety, including studies using electronic health records, claims data, registries, and other observational sourcesCausal inference with real-world data, including treatment effect estimation, target trial emulation, instrumental variables, and methods for handling confounding and selection biasCausal machine learning, including counterfactual prediction, heterogeneous treatment effects, and robustness under dataset shiftMultimodal and multi-source integration, linking clinical outcomes with biological signals, omics data, and other evidence streamsDrug safety and pharmacovigilance, including adverse event detection, drug-drug interactions, polypharmacy risks, and signal validation using AI and RWEEvaluation, robustness, and uncertainty, including sensitivity analysis, transportability, uncertainty quantification, and reproducibility practices Topic editors: Yi Guo, Laszlo Barna Iantovics Source: https://www.frontiersin.org/research-topics/82850/ai-and-real-world-evidence-for-drug-discovery-repurposing-and-safety