Machine Learning-Assisted Dual-Marker Detection in Serum Small Extracellular Vesicles for the Diagnosis and Prognosis Prediction of Non-Small Cell Lung Cancer.

Small extracellular vesicles (sEVs) carry molecular information from their source cells and are desired biomarkers for cancer diagnosis. We establish a machine learning-assisted dual-marker detection method to analyze the expression of epidermal growth factor receptor (EGFR) and C-X-C chemokine receptor 4 (CXCR4) in serum sEVs for the diagnosis and prognosis prediction of non-small cell lung cancer (NSCLC). We find that the serum sEV EGFR and CXCR4 are significantly higher in advanced stage NSCLC (A/NSCLC) patients compared to early stage NSCLC (E/NSCLC) patients and the healthy donors (HDs). A receiver operating characteristic curve (ROC) analysis demonstrates that the combination of EGFR and CXCR4 in serum sEVs as an efficient diagnostic index and malignant degree indicator for NSCLC. Machine learning further shows a diagnostic accuracy of 97.4% for the training cohort and 91.7% for the validation cohort based on the combinational marker. Moreover, this machine leaning-assisted serum sEV analysis successfully predicts the possibility of tumor relapse in three NSCLC patients by comparing their serum sEVs before and three days after surgery. This study provides an intelligent serum sEV-based assay for the diagnosis and prognosis prediction of NSCLC, and will benefit the precision management of NSCLC.