Systematic mapping of artificial intelligence and big data architectures for antidiabetic drug discovery from traditional medical systems

Abstract

Abstract Background Ethnomedical knowledge has contributed to the discovery and development of several contemporary antidiabetic agents. However, the extensive global diversity of medicinal plants, phytochemicals, and traditional therapeutic systems creates a substantial challenge for systematic prioritization. This review was conducted to map the application of artificial intelligence (AI) and big data approaches to support evidence-based prioritization in antidiabetic ethnopharmacology. Methods A systematic mapping review was conducted and reported in accordance with PRISMA 2020. PubMed and Scopus were searched using predefined terms related to diabetes, medicinal plants or traditional medicine, and AI or big-data computational methods. Eligible studies were synthesized using descriptive statistics to characterize the distribution of AI methodologies, data sources, computational workflows, ethnopharmacological integration, and validation levels. Results Twenty-nine studies met the inclusion criteria. Machine learning was the most frequently applied approach (25%), mainly for compound recognition, bioactivity prediction, and candidate prioritization. Many studies used dual-focus computational architectures, in which prediction-oriented AI methods, including machine learning or deep learning, were combined with mechanistic frameworks such as network pharmacology and molecular modeling (38%). Validation was mainly restricted to in silico analyses. Ethnopharmacological integration was commonly mediated by secondary database proxies, such as herb lists, indications, and compound–target annotations, rather than by structured modeling of primary ethnomedicinal practice. Conclusions AI and big-data methods are being used to reshape selected stages of antidiabetic ethnopharmacology, particularly candidate prioritization, target prediction, and network-based mechanism generation. However, most evidence remains hypothesis-generating.