Machine learning-assisted assessment of extracellular vesicles can monitor cellular rejection after heart transplant
Abstract Background Heart transplant rejection, particularly acute cellular rejection (ACR), remains a critical post-operative concern, despite declining incidence rates. Current diagnostic standards rely on invasive endomyocardial biopsy, which presents limitations in sensitivity and reproducibilit...
Saved in:
| Main Authors: | , , , , , , , , , , , , , , , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Nature Portfolio
2025-07-01
|
| Series: | Communications Medicine |
| Online Access: | https://doi.org/10.1038/s43856-025-00999-0 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Abstract Background Heart transplant rejection, particularly acute cellular rejection (ACR), remains a critical post-operative concern, despite declining incidence rates. Current diagnostic standards rely on invasive endomyocardial biopsy, which presents limitations in sensitivity and reproducibility. There is an unmet need for noninvasive, accurate biomarkers that can detect and monitor rejection. This study aims to evaluate whether extracellular vesicle (EV) surface antigens, analyzed through flow cytometry and interpreted with artificial intelligence (AI), can serve as reliable biomarkers for ACR detection and monitoring in heart transplant recipients. Methods We conducted a prospective longitudinal cohort study involving 24 heart transplant recipients over a median follow-up of 303 days. A total of 285 blood samples were analyzed for EV surface antigens exploiting two flow cytometry-based protocols. An adaptive AI model (random forest regressor) was developed to interpret EV antigen profiles, dynamically calibrating thresholds per patient. Results Here we show that 14 EV surface antigens progressively increase with ACR severity. These changes are evident even before histological diagnosis. The AI model achieves an accuracy of 93.3% at leave-one-out testing (AUC 0.968), and 78.9% at validation in an independent cohort (AUC 0.832), with high specificity and negative predictive value. EV profiling outperforms conventional biochemical markers and provides anticipatory insight into rejection dynamics. Conclusions EV profiling, enhanced by patient-specific AI modeling, offers a powerful noninvasive method for early detection and monitoring of ACR. This approach holds the potential to reduce reliance on biopsies and tailor immunosuppressive strategies more precisely. |
|---|---|
| ISSN: | 2730-664X |