Machine learning multiscale collective cell dynamics: From single-cell characterization to multicellular monolayer...

Area of Science:

• Computational Biology
• Biophysics
• Machine Learning in Biology

Background:

• Collective cell dynamics are crucial for physiological processes like development and disease.
• Traditional models struggle with unmeasurable parameters and complex active physics across scales.
• Accurate multiscale modeling of cell behavior is essential for understanding biological systems.

Purpose of the Study:

• To develop a novel scale-adaptive hybrid machine learning (ML) framework for modeling collective cell dynamics.
• To address limitations of traditional physics-based models in capturing single-cell and multicellular behaviors.
• To integrate physics-guided and physics-agnostic ML methods for improved predictive power.

Main Methods:

• Developed a hybrid ML framework combining physics-guided and physics-agnostic approaches.
• Physics-guided ML inferred unmeasurable parameters from experimental data for single-cell analysis.
• Physics-agnostic ML modeled multicellular dynamics directly from historical state sequences.

Main Results:

• The physics-guided ML method accurately characterized single-cell features but had limitations in predicting multicellular dynamics.
• The physics-agnostic ML method robustly predicted coarse-grained multicellular behaviors like density oscillations.
• The hybrid framework leveraged the strengths of each method, mitigating individual limitations for multiscale modeling.

Conclusions:

• The scale-adaptive hybrid ML framework offers a versatile computational paradigm for multiscale collective cell dynamics.
• This approach bridges the gap between theoretical modeling and experimental observations in biology.
• The framework has broad potential applications in diverse physiological and pathological contexts.