SERS spectroscopy with machine learning to analyze human plasma derived sEVs for coronary artery disease diagnosis and prognosis

FIGURE 1. Schematic illustration of ML‐assisted sEV analysis for CAD diagnosis. (a) Isolation of sEVs from the human plasma of patients with different CAD stages, including HC, SP, NSTEMI and STEMI stages, and collection of spectroscopic data of plasma‐derived sEVs by SERS. (b) Overview of ML‐assisted plasma sEV classification and CAD diagnosis using sEVs SERS signal patterns. This illustration is created by BioRender.com with an authorized license

FIGURE 2. Isolation and characterizations of sEV from human plasma. (a) Scheme of isolation procedure of sEVs from human plasma. (b) AFM images of sEVs. (c, d) The size and concentration of sEVs by NTA test (n = 11, *p < 0.05). (e) Western blot analysis of biomarker proteins on sEVs

FIGURE 3. Raman spectra and QDA analysis. (a) Averaged Raman spectra for four stages of CAD (HC, SP, NSTEMI, STEMI); (b) The PCA plot colored by QDA classification results of the total 775 Raman spectra from four stages of CAD. (c) Heat map of the prediction possibility (range 0–1) by QDA for four stages of CAD. (d) ROC curves for each representation stage and their AUC values. (e) Prediction accuracy of test sets performed in QDA, SVM, KNN, ANN, and XGB in 50 test rounds by randomly splitting 90% of data into a training set and 10% into a test set

FIGURE 4. SVM analysis. (a) A PCA plot colored by SVM classification results of the total 775 Raman spectra from four stages of CAD. (b) Heat map of the prediction possibility (range 0–1) by SVM for four stages of CAD. (c) ROC curves for each representation stage and their AUC values. (d) Decision boundaries of SVM projected in 2D PCA plot (PC1 vs PC2). (e) Decision boundaries of SVM projected in the 2D PCA plot (PC2 vs PC3). (Black: HC; Red: SP; Green: STEMI; Blue: NSTEMI)