Machine learning interpretable models of cell mechanics from protein images

Area of Science:

• Cellular Mechanobiology
• Computational Biology
• Biophysics

Background:

• Cellular form and function arise from complex mechanochemical systems.
• A systematic strategy to infer large-scale physical properties from molecular components is lacking.
• This knowledge gap hinders understanding of cell adhesion and migration.

Purpose of the Study:

• To develop a data-driven modeling pipeline for inferring mechanical behavior in adherent cells.
• To establish a method for predicting cellular forces from images of cytoskeletal proteins.
• To integrate neural networks into predictive models for cell biology.

Main Methods:

• Trained neural networks to predict cellular forces from images of cytoskeletal proteins.
• Utilized images of single focal adhesion (FA) proteins, like zyxin, to predict forces.
• Developed physics-constrained and agnostic data-driven continuum models of cellular forces.

Main Results:

• Experimental images of single FA proteins are sufficient to predict cellular forces.
• The predictive models generalize to unseen biological regimes.
• Both modeling approaches revealed that cellular forces are encoded by two distinct length scales.

Conclusions:

• A data-driven pipeline can successfully predict cellular mechanical behavior from molecular imaging.
• Neural network integration offers a powerful approach for modeling cell biology.
• The findings provide insights into how cellular forces are encoded at different length scales.