Deep Learning-Based Classification of NSCLC-Derived Extracellular Vesicles Using AFM Nanomechanical Signatures

Area of Science:

• Biophysics
• Cancer Biology
• Nanotechnology

Background:

• Nonsmall cell lung cancer (NSCLC) is a major cause of cancer mortality.
• Liquid biopsy offers a noninvasive approach for NSCLC diagnostics.
• Extracellular vesicles (EVs) are crucial tumor microenvironment messengers, but their characterization is challenging.

Purpose of the Study:

• To characterize nanomechanical properties of EVs from different NSCLC subtypes using atomic force microscopy (AFM).
• To develop a deep learning model for enhanced classification of NSCLC-derived EVs.
• To assess the potential of AFM and deep learning for advancing liquid biopsy in NSCLC.

Main Methods:

• AFM was employed to analyze the stiffness and nanomechanical properties of EVs from NSCLC cell lines (A549, PC9, PC9/GR) and a control cell line (BEAS-2B).
• A DenseNet-based deep learning model was developed to integrate nanomechanical and morphological features from AFM images for EV classification.
• Diagnostic performance was evaluated using Area Under the Curve (AUC) and classification accuracy.

Main Results:

• EVs derived from A549 cells (KRAS mutant) exhibited significantly higher stiffness compared to other cell types, potentially due to lipid alterations.
• EVs from EGFR-mutant NSCLC cell lines (PC9, PC9/GR) displayed overlapping nanomechanical properties.
• The deep learning model achieved an AUC of 0.92, with 96% accuracy in classifying A549-derived EVs.

Conclusions:

• This study demonstrates the first nanomechanical classification of NSCLC-derived EVs using AFM and deep learning.
• The integration of nanomechanical and morphological data significantly enhances diagnostic performance for liquid biopsy.
• Deep learning-enhanced AFM analysis holds promise for precision diagnostics in NSCLC, with future validation in clinical samples needed.