Artificial intelligence for extracellular vesicle engineering and therapeutic design

Extracellular vesicles (EVs) are promising platforms for next-generation drug delivery and regenerative therapeutics, yet biological heterogeneity, limited targeting specificity, variable cargo composition, and continued reliance on empirical optimization hinder clinical translation. Artificial intelligence (AI) offers new opportunities to make EV engineering more predictive and data-driven. Building on prior work in AI-assisted EV delivery and emerging virtual-EV modeling, this review presents an evidence-aware, multiscale framework that integrates parental-cell programming, vesicle design, recipient-system modeling, and experimental feedback. At the source level, AI virtual cells (AIVCs) may help predict how perturbations to parental cells influence EV biogenesis and composition. At the vesicle level, AI-guided approaches may support cargo prioritization, surface engineering, and loading optimization. At the recipient level, digital twin models could help simulate EV uptake, intracellular trafficking, microenvironmental interactions, and downstream therapeutic responses. These components may ultimately be integrated with experimental validation into iterative design-test-refinement workflows. Importantly, this review distinguishes direct EV-specific evidence from technologies transferable from adjacent fields and concepts that remain prospective. By defining these evidence boundaries and linking them to key challenges in data quality, model interpretability, manufacturing, and validation, this framework provides a practical roadmap for advancing AI-enabled EV therapeutics from conceptual modeling to testable, translationally relevant engineering strategies.