Prospective Artificial Intelligence for Drug Discovery: From Prediction to Experimental Validation

Artificial intelligence (AI) and machine learning are increasingly applied to virtual screening, molecular property prediction, generative design, retrosynthesis, ADME/Tox prediction and multiparameter optimisation. Yet many methods are assessed mainly on historical datasets that may not reflect distribution shifts, synthetic constraints, assay variability, limited data or the practical consequences of discovery decisions. Advances in make-on-demand libraries, rapid synthesis, high-throughput experimentation, biological screening and integrated data infrastructures are creating new opportunities to test model-generated hypotheses prospectively. The field is therefore moving beyond whether AI can reproduce existing data to whether it can guide useful new experiments and improve subsequent decisions. Prospective evaluation can reveal where models generalise or fail, and how experimental evidence can inform future predictions.This Research Topic aims to advance prospective validation as a practical standard of evidence for AI-enabled drug discovery. We seek research on AI or machine-learning systems that inform molecular design, compound selection, prioritisation or experimental decisions, and on approaches to evaluate these systems against appropriate baselines. Contributions may investigate how uncertainty guides experimental selection and how experimental feedback refines predictions. By bringing together computational methods, prospective studies, iterative design-make-test-learn workflows, benchmarks and critical perspectives, this Research Topic aims to advance rigorous, reproducible and decision-relevant AI in medicinal chemistry and drug discovery.To gather further insights into prospective AI for drug discovery, this Research Topic will cover computational and experimental workflows in which model outputs inform discovery decisions. Computational studies are welcome; prospective experimental validation and integration with experimental workflows are encouraged as important means of assessing real-world utility, without being a requirement for every contribution. We welcome articles addressing, but not limited to, the following themes:&#8226 Prospective virtual screening and predictive modelling, including drug–target interactions, binding affinity and drug response&#8226 Generative AI for target-aware, structure-guided and synthesis-aware molecular design&#8226 AI-assisted medicinal chemistry, molecular generation and multi-objective optimisation&#8226 Active learning, Bayesian optimisation, uncertainty quantification and out-of-distribution decision-making&#8226 Multi-omics-guided, multimodal and context-aware AI for biologically informed or precision-oriented discovery&#8226 Iterative design–make–test–learn workflows and experimental feedback for model refinement&#8226 Prospective benchmarking, reproducibility and transparent reporting of negative or unsuccessful results&#8226 Human-in-the-loop AI, laboratory automation and data infrastructure supporting discovery workflows Topic editors: Mohamed S. H. Salem, Pir Masoom Shah Source: https://www.frontiersin.org/research-topics/86575/prospective-artificial-intelligence-for-drug-discovery-from-prediction-to-experimental-validation