Integrating Artificial Intelligence and Machine Learning into Natural Product Discovery for Human Health from Ethnopharmacological Knowledge to Molecular Mechanisms

Abstract

Background: Ethnopharmacology has accumulated millennia of traditional medicinal knowledge, yet natural product discovery remains constrained by low screening efficiency, delayed mechanistic elucidation, and fragmented translation from traditional knowledge to clinical application. Artificial intelligence (AI) and machine learning (ML) offer transformative tools for systematically mining traditional knowledge, predicting bioactivity, and deconstructing multi-target mechanisms.
Methods: This review systematically integrates representative research on AI/ML in natural product discovery from 2020–2026, covering chemical language models, graph neural networks, generative AI, network pharmacology, molecular docking, and dynamics simulation. Based on 76 references, a five-stage integrated framework is constructed: knowledge digitization, activity prediction, mechanism deconstruction, and translational assessment.
Results: AI/ML has demonstrated significant advantages across ethnopharmacological data mining, virtual screening of natural product libraries, ADMET prediction, multi-target mechanism network analysis, and personalized ethnomedicine. Generative models (e.g., MolBERT, DrugGPT, GENTRL) can complete the cycle from generation to lead optimization within 46 days. The integration of network pharmacology with HTS² enables coupling of computational prediction and transcriptomic validation. However, the field faces systemic challenges including data fragmentation, benchmark absence, insufficient interpretability, and lagging ethical governance.
Conclusion: The deep integration of AI/ML with ethnopharmacology requires a shift from "tool stacking" to "paradigm reconstruction." Key priorities include building standardized digital infrastructure for traditional knowledge, developing specialized models for natural product chemical space, and embedding mechanistic validation into computational workflows. Future directions include multimodal foundation models, closed-loop experimental platforms, and community-engaged ethical frameworks.