Microfluidic Biochip-Based Multiplexed Profiling of Small Extracellular Vesicles Proteins Integrated with Machine Learning for Early Disease Diagnosis

Abstract

Accurate early diagnosis is essential for preventing diseases and improving cure and survival rates. There are no reliable early-diagnosis biomarkers for most major diseases. Here, esophageal squamous cell carcinoma (ESCC) is used as a disease model to develop a platform for detecting a panel of proteomic biomarkers for accurate early diagnosis by integrating a barcode immunoassay biochip with machine learning. The biochip captures small extracellular vesicles (EVs) from serum, lyses them in situ, and quantifies multiple proteins, including membrane and internal proteins of EVs. It is utilized to test 273 clinical samples across multiple centers. The validation sets are then analyzed using machine learning, resulting in a precise diagnostic model for ESCC. This model, based on nine diagnostic protein biomarkers identified through mass spectrometry analysis of differentially expressed proteins, achieves an accuracy of 91.0% in external validation, with a 90.8% accuracy in detecting early-stage ESCC. These results significantly surpass the accuracy (only 14.4%) of the currently used biomarker for squamous cell carcinoma. Thus, integrating extracellular vesicles protein analysis with machine learning presents can identify ESCC patients. The developed extracellular vesicles analysis platform offers a promising tool for the clinical application of multi-biomarker detection methods, advancing the early diagnosis of ESCC.