Leveraging agent-based models and deep reinforcement learning to predict taxis in cell migration.

Area of Science:

• Computational biology
• Systems biology
• Bioinformatics

Background:

• Traditional rule-based agent-based models (ABM) often require explicit definition of cellular behaviors.
• Understanding how cells sense and respond to environmental cues is crucial in biology.

Purpose of the Study:

• To develop a novel computational framework integrating agent-based modeling (ABM) with reinforcement learning (RL) for predicting cellular responses to environmental signals.
• To enable models to learn cellular behavior directly from experimental data, avoiding predefined rules.

Main Methods:

• Combined Agent-Based Modeling (ABM) with the Double Deep Q-Network (DDQN) algorithm, a type of reinforcement learning (RL).
• The framework captures the transduction of environmental cues into biological responses directly from experimental observations.
• Applied the model to analyze barotactic cell migration data from microfluidic experiments with varying pressure gradients.

Main Results:

• The model successfully predicted dynamic, environment-dependent cell behavior.
• Demonstrated the framework's ability to generalize across different microfluidic device geometries and pressure configurations.
• Showcased the direct learning of cellular responses to environmental cues without explicit behavioral predefinition.

Conclusions:

• The proposed ABM-RL framework offers a scalable and flexible alternative to traditional rule-based ABM for modeling cellular behavior.
• This approach provides a novel direction for understanding how cells sense and transduce environmental cues into adaptive biological behaviors.
• The model's success in predicting barotactic cell migration highlights its potential for diverse biological applications.