Artificial intelligence for natural product drug discovery and development: current landscape, applications, and future directions

Artificial intelligence accelerates natural product discovery in oncology, infection, inflammation, and neuroprotection by enabling activity prediction, mechanism inference, and prioritization. These approaches include tree ensembles, graph neural networks, and self-supervised molecular embeddings for mixtures, isolated metabolites, and peptide analogs, while network pharmacology models herb–ingredient–target–pathway graphs to propose synergistic effects. Operational multi-omics gates (transcriptomic signature reversal, proteome-scale target engagement, and untargeted metabolomics with feature-based molecular networking) move ranked candidates into reproducible validations. Population-level analyses map formulation-derived ingredient–target signatures to clinical outcomes, and large language models are beginning to standardize herbal prescriptions and their curation. Persistent barriers include mixture and batch variability, incomplete provenance, small and imbalanced datasets, domain shift, off-target liability, limited interpretability, and bias. Practical solutions include minimal information for AI on natural product metadata for provenance and safety, scaffold and time-split benchmarks with cross-lab replication, uncertainty and applicability-domain gating, mechanistic add-back experiments, constrained generative and semi-synthetic design, micro-physiological systems with digital twins, and provenance-aware pharmacovigilance under evolving FDA, EMA, and WHO expectations. This review integrates these advances, evidence gaps, and governance requirements into a roadmap for the mechanistically grounded, prospectively validated translation of AI in natural product research. • AI tools accelerate natural product–based drug discovery and development. • Machine learning and deep learning models predict anticancer, anti-inflammatory, and antimicrobial actions. • Several AI-predicted natural compounds were validated in vitro, confirming translational potential. • Key challenges include small datasets, data imbalance, and limited experimental validation.