Artificial Intelligence Artificial intelligence-guided biosynthesis and retrosynthesis in pharmacognosy: Toward the synthesis-oriented discovery of natural-product therapeutics

Abstract

Natural products continue to be one of the most important sources of therapeutic scaffolds, but turning them into real drug candidates is rarely straightforward. Their development is often slowed by structural complexity, low abundance in nature, difficult isolation, incomplete knowledge of their biosynthetic origins, and the challenge of achieving stereoselective synthesis. This review looks at how artificial intelligence is helping move pharmacognosy toward a more synthesis-oriented discipline by linking ethnobotanical knowledge, multi-omic data, biosynthetic pathway prediction, retrosynthetic planning, and medicinal chemistry optimization. A key focus is placed on AI-driven platforms that can reconstruct natural-product biosynthesis and support synthetic or semi-synthetic route design for complex metabolites. Tools such as BioNavi-NP, graph-sequence enhanced transformers, NAG2G, RSGPT, RetroExplainer, and human-in-the-loop systems like DeepRetro show how transformer-based, graph-based, and large language model-assisted approaches can predict plausible precursors, retain molecular topology, suggest multi-step disconnections, and explore broad reaction spaces. These methods are especially useful for natural products with dense stereochemistry, unusual ring systems, multifunctional scaffolds, and enzyme-guided biosynthetic logic, features that conventional retrosynthetic approaches may not always capture well. Beyond route prediction, AI can help prioritize biosynthetic genes, support the optimization of scarce plant-derived compounds, and guide the design of more drug-like analogues with improved potency, selectivity, pharmacokinetic behavior, and synthetic accessibility. Still, several limitations remain. These include the lack of plant-specific reaction datasets, weak prediction of reaction conditions, incomplete modeling of stereochemistry and regioselectivity, benchmark-related problems, and the ongoing need for expert chemical validation. Overall, AI should not be seen as a replacement for pharmacognosists, synthetic chemists, or medicinal chemists. It is better understood as a decision-support layer that connects biodiversity, traditional knowledge, biosynthetic logic, and experimental synthesis. With more transparent, plant-specific, and experimentally validated AI systems, the discovery and responsible development of natural-product-based therapeutics could become faster and more reliable.