fig5

Figure 5. Closed-loop AI-driven EV engineering pipeline. The diagram presents an adaptive optimization cycle for EV therapeutic development. Prioritized EV candidates are first evaluated through computational prediction and prioritization, in which predicted performance and model uncertainty may help identify candidates for experimental testing. Selected vesicles are then produced and assessed through functional assays, and the resulting readouts can update the model and rerank candidates. Experimental feedback can subsequently inform model refinement, revised candidate ranking, and the selection of promising EV candidates for subsequent testing cycles. Created in BioRender. AI: Artificial intelligence; EV: extracellular vesicle.