Host Cell Prediction of Exosomes Using Morphological Features on Solid Surfaces Analyzed by Machine Learning.

Area of Science:

• Biophysics
• Nanotechnology
• Cancer Research

Background:

• Exosomes, extracellular nanovesicles found in body fluids, carry host cell information.
• Exosome analysis holds potential for early cancer diagnosis due to their cell-specific markers.
• Characterizing exosome properties is crucial for understanding their diagnostic capabilities.

Purpose of the Study:

• To predict the host cancer cells of origin by analyzing exosome physical characteristics.
• To develop a machine learning model for classifying exosomes based on multidimensional AFM data.
• To investigate the influence of different substrates on exosome analysis and prediction accuracy.

Main Methods:

• Atomic Force Microscopy (AFM) was employed to immobilize and image exosomes on various substrates (SiO2/Si, modified SiO2/Si, TiO2).
• Multidimensional feature data (size, shape, deformation) were extracted from AFM images, creating 14-dimensional feature vectors.
• Support Vector Machine (SVM) learning and Principal Component Analysis (PCA) were utilized for classification and interpretation of exosome particle data.

Main Results:

• High prediction accuracies were achieved: 85.2% for two-class SVM and 82.6% for three-class SVM.
• The study identified substrate-dependent variations in exosome analysis.
• PCA analysis revealed that the aspect ratio of exosomes, decreasing with volume, significantly impacts prediction accuracy and substrate dependence.

Conclusions:

• AFM combined with SVM and PCA provides a robust method for predicting exosome host cells.
• Exosome physical properties, particularly aspect ratio and volume, are key discriminators for host cell identification.
• The choice of substrate significantly influences the accuracy of exosome-based cancer diagnostics.