Aims
Extracellular vesicles (EVs) are lipid bilayer-bound nanoparticles secreted by all cell types. They serve critical roles in intercellular communication by transporting bioactive molecules, including proteins, lipids, nucleic acids, and metabolites. These molecular contents reflect the physiological or pathological state of the parent cell and can influence gene expression, signalling pathways, and cellular behaviour in recipient cells. Based on size and biogenesis, EVs are broadly categorized into small and large subtypes. Due to their stability and molecular specificity, EVs have emerged as valuable tools in diagnostics, therapeutic delivery, and disease monitoring. However, determining the cellular origin of EVs in complex biological fluids such as blood, plasma, and urine remains a major challenge. Current experimental methods often lack resolution, scalability, or affordability.
Methodology
To address this challenge, we developed EV Barcode, an open-source web platform that employs machine learning to infer the cellular source of EVs based on omics-derived cargo profiles. The predictive model was trained on over five million single-cell transcriptomes from more than one thousand human tissue samples obtained from public datasets such as the Gene Expression Omnibus and the Single Cell Portal. Model development included feature selection, hyperparameter optimization, and ensemble learning, validated through rigorous cross-validation and performance benchmarking.
Results and Conclusion
EV Barcode generates statistically supported predictions of EV origin at the cell and tissue level. The platform enables interactive visualizations to examine cell-type contributions and their dynamic variation across experimental conditions. EV Barcode offers a powerful resource for advancing EV research and accelerating clinical translation.