Artificial Intelligence in Drug Discovery: A Theoretical Framework of Target Identification, Molecular Design and the Translation Gap for Diseases of the Global South

Abstract

Artificial intelligence (AI) has compressed stages of drug discovery that once required years. It predicts protein structures, screens virtual libraries, generates candidate molecules and forecasts their properties, and the first AI-discovered molecules have entered clinical trials. This paper asks whether that acceleration will reach the diseases that burden the Global South, drawing on the pharmaceutical-sciences literature on AI-driven discovery, the economics of pharmaceutical innovation and neglected diseases, and recent research on machine learning for target identification, de novo molecular design and property prediction. It distinguishes three stages, target identification, molecular design and translation to the clinic, and argues that AI has transformed the first two while leaving the third constrained by the factors that constrained it before: the cost of trials, the absence of commercial incentive for low-margin diseases, and weak research and regulatory infrastructure where those diseases occur. On this basis the paper advances the view that AI's effect on the global distribution of new medicines depends on three variables: who owns the models, whose disease targets and chemical spaces the models are trained upon, and whether reductions in the cost of discovery are carried through into the development of unprofitable indications. It then considers the conditions under which African pharmaceutical research institutions might participate in AI-driven discovery for malaria, neglected tropical diseases and antimicrobial resistance, rather than remaining downstream consumers of its products. The paper derives a set of propositions about when AI-driven discovery reaches neglected diseases, and argues that participation on terms that build local capacity, rather than consumption of tools designed elsewhere, is what would allow African institutions to shape the chemical spaces and disease targets on which the models are trained.