AI-Guided Patient-Derived Endothelial Modeling Identifies a Reversible Moyamoya-Specific miRNA Deficiency Corrected by Stem Cell-Extracellular Vesicles

Clinical Snapshot

Key Findings

• P Value: Not specified

• Effect Size: Majority of depleted miRNAs restored to normal levels

• Primary Outcome: Significant depletion of angiogenesis-related miRNAs in MMD patients

• Nnt Or Sensitivity: AUROC improved from 0.51 to 0.72 with enhanced prediction model

• Confidence Interval: Not reported

Clinical Application

Currently experimental; requires specialized iPSC facilities and MSC-EV production Would require TGA approval for clinical translation; limited Moyamoya expertise outside major centres; potential PBS considerations for future cell therapy applications Moyamoya disease patients, particularly those with endothelial dysfunction

Abstract

Moyamoya disease (MMD) is a progressive cerebrovascular disorder characterized by abnormal vascular remodeling and impaired angiogenesis. We profiled plasma extracellular vesicle (EV)-derived microRNAs (miRNAs) from MMD patients and healthy controls and generated patient-specific induced pluripotent stem cell-derived endothelial cells (iPSC-ECs) to investigate disease-associated molecular alterations and therapeutic reversibility. Both plasma EVs and iPSC-ECs from MMD patients showed significant depletion of angiogenesis-related miRNAs. Small RNA sequencing data revealed mesenchymal stem cell-derived EVs (MSC-EVs) contain a miRNA repertoire that substantially overlaps with miRNAs deficient in MMD. Treatment of MMD iPSC-ECs with MSC-EVs restored the majority of the depleted miRNAs to normal levels, normalized predicted target mRNA expression, and rescued angiogenic function in a dose-dependent manner, as demonstrated by tube formation assays. To improve identification of biologically relevant miRNA-mRNA interactions, we developed an artificial intelligence-based framework integrating experimentally validated databases, endothelial-focused literature mining, and large language model (LLM)-assisted mechanistic scoring. Incorporation of LLM-derived contextual features modestly improved prediction performance compared with conventional sequence-based models (baseline AUROC ~ 0.51; LLM-enhanced AUROC ~ 0.72), suggesting that literature-informed biological context provides meaningful signal beyond structural target features alone. Furthermore, our findings demonstrate that MSC-EV therapy can substantially reverse molecular and functional endothelial deficits. Integrated profiling and AI-driven interaction mapping together establish a hypothesis-generating framework ("Exo-Courier") for biologically informed prioritization of EV-based therapeutic targets in vascular disorders.

References

1. 1.Bang, J. H., Kim, J. E., Kim, E. H., Yang, H., Jang, J., & Bang, O. Y. (2026). AI-Guided Patient-Derived Endothelial Modeling Identifies a Reversible Moyamoya-Specific miRNA Deficiency Corrected by Stem Cell-Extracellular Vesicles. Translational Stroke Research. https://doi.org/10.1016/j.ymeth.2024.01.011