Integration of Raman tweezers and machine learning for label-free single-cell characterization of endometriosis cells

Abstract

Endometriosis occurs when endometrial tissue grows outside the uterus, affecting millions of women worldwide. Despite extensive research, its cellular mechanisms remain unclear, complicating both diagnosis and treatment. This study presents the development and validation of a Raman tweezers platform that combines optical trapping and Raman spectroscopy for label-free biochemical profiling of single endometriosis-derived VK2/E6E7 epithelial cells. To assess discriminatory capacity, VK2/E6E7 spectra were compared against the epithelial cancer cell line A549. The Raman tweezers system was calibrated with polystyrene beads, and spectral data were preprocessed using the self-supervised deep learning model (RSPSSL) to ensure reproducible single-cell measurements. The Raman spectrum of VK2/E6E7 cells displays characteristic peaks corresponding to lipids, collagen (418, 606, 1312, and 1447 cm⁻¹), proteins (538, 938, 998, 1258, and 1447 cm⁻¹), and nucleic acids (737, 1093, 1187, and 1258 cm⁻¹). Random Forest and XGBoost for classifying VK2/E6E7 and A549 cells achieved over 80% accuracy without signs of overfitting. SHAP (SHapley Additive exPlanations) analysis highlighted lower lipid, amino acid, and amide III signals alongside higher saccharide signals as key drivers of cell differentiation. This is the first study to apply Raman tweezers for single-cell analysis of endometriosis cells, integrating deep-learning preprocessing and explainable machine learning. It offers a promising approach for probing endometriosis pathophysiology and supporting a less invasive diagnostic strategy.