Traction force microscopy by deep learning.

Area of Science:

• Mechanobiology
• Cellular mechanics
• Biophysics

Background:

• Cells exert and sense forces, crucial for mechanobiology.
• Traction Force Microscopy (TFM) maps cell-generated forces but faces accuracy/resolution limits due to ill-posed mathematical problems.
• Conventional TFM requires compromises in accuracy and/or resolution.

Purpose of the Study:

• To develop a deep learning-based approach for Traction Force Microscopy (TFM).
• To improve the accuracy and applicability of TFM for analyzing cell-generated forces.
• To overcome limitations of conventional TFM methods.

Main Methods:

• Applied a modified neural network for predicting stress vector fields from displacement data.
• Utilized a mathematical model of cell migration to generate extensive simulated data for training and testing.
• Trained a deep learning model using simulated cell migration stresses and displacements.

Main Results:

• Deep learning-TFM achieved higher accuracy compared to tested conventional TFM implementations.
• Results from deep learning-TFM closely resemble those from conventional TFM.
• The trained neural network demonstrated applicability across various conditions (cell size, shape, substrate stiffness, traction output).

Conclusions:

• Deep learning offers a powerful alternative for TFM, enhancing accuracy and resolution.
• This method provides a more robust tool for characterizing mechanical interactions between cells and their environment.
• The developed deep learning-TFM is a versatile and appealing alternative to conventional techniques.