Quantifying the Dynamics of Protein Self-Organization Using Deep Learning Analysis of Atomic Force Microscopy Data

Area of Science:

• Surface science
• Biophysics
• Materials science

Background:

• Protein self-assembly on inorganic surfaces is crucial for various applications.
• Understanding the dynamics and geometric patterns of this process is challenging.
• Existing methods lack the resolution to capture fine details of particle behavior.

Purpose of the Study:

• To visualize and analyze the dynamics of protein self-assembly on inorganic surfaces.
• To develop a computational workflow for analyzing complex self-organization processes.
• To identify distinct phases and transitions in protein self-assembly.

Main Methods:

• High-speed atomic force microscopy (HS-AFM) for visualization.
• Unsupervised linear unmixing for analyzing macroscopic descriptors (2D FFT, correlation, pair distribution functions).
• Deep learning (DL)-based workflow for particle dynamics and local geometry analysis.

Main Results:

• Visualized protein self-assembly dynamics and geometric patterns.
• Demonstrated the presence of static ordered and dynamic disordered phases.
• Established a DL workflow to analyze particle dynamics and local geometry evolution.
• Separated particle behaviors and identified transitions using DL feature extraction and mixture modeling.

Conclusions:

• The developed workflow enables analysis of self-organization in complex systems from observational data.
• Provides fundamental insights into the mechanisms of protein self-assembly.
• Highlights the utility of deep learning in analyzing dynamic nanoscale phenomena.