Machine Learning-Based Identification and Validation of Plasma Exosomal miRNAs as Diagnostic Biomarkers for Early-Stage Nasopharyngeal Carcinoma

Background: Exosomal microRNAs (miRNAs) have been implicated in cancer pathophysiology, but their role in early-stage nasopharyngeal carcinoma (NPC) remains unclear. This study employed machine learning models to identify and validate key exosomal miRNAs associated with early-stage NPC.Methods: A total of 8 patients with early-stage NPC and 9 negative controls were enrolled. Exosomes were isolated using the exoEasy Maxi Kit and characterized by transmission electron microscopy (TEM) and Western blot analysis. Small RNA sequencing was performed to compare miRNA expression profiles between the two groups. The GSE70970 dataset was integrated with sequencing data for machine learning-based diagnostic modeling. The development set was used for model training, followed by optimization using the test set. Key miRNAs identified by the machine learning model were further validated via PCR-based quantitative assays. Finally, target gene prediction and validation were conducted for these key miRNAs. Results: TEM revealed that the isolated exosomes exhibited typical round or oval-shaped bilayer membrane structures. Western blot confirmed the expression of exosomal marker proteins CD63 and CD81. High-throughput sequencing identified 43 differentially expressed exosomal miRNAs between the two groups. After integrating the GSE70970 dataset, 14 commonly upregulated differentially expressed miRNAs were selected for Random Forest analysis. The dataset was divided into development and test sets at a 50% ratio to construct the NPC diagnostic model. The top 5 miRNAs (hsa-miR-141-5p, hsa-miR-511-5p, hsa-miR-23a-5p, hsa-miR-579-3p, hsa-miR-548o-3p) ranked by area under the curve (AUC) were used to plot receiver operating characteristic (ROC) curves. When all 5 miRNAs were included, the AUC reached 1.0 in the development set and 0.905 in the test set. RT-qPCR validation demonstrated a significant difference in miR-579-3p expression levels between the two groups (P < 0.05). Target gene prediction and validation indicated that the associated genes were primarily enriched in the Wnt signaling pathway.Conclusion: The combination of five plasma exosomal miRNAs exhibits strong predictive value for early-stage NPC. The Wnt signaling pathway may be involved in the early development of NPC, and whether exosomal miRNAs participate in NPC pathophysiology by regulating chemoattractant signaling pathways warrants further investigation.