Microswimmers learning chemotaxis with genetic algorithms

Area of Science:

• Biophysics
• Computational Biology
• Artificial Intelligence

Background:

• Microorganisms and mammalian cells utilize nonreciprocal body deformations for locomotion in viscous fluids.
• Chemotaxis, the directed movement towards nutrients, requires sophisticated adaptation of swimming gaits.

Purpose of the Study:

• To develop a computational model for autonomous shape adaptation in microswimmers.
• To investigate the role of artificial neural networks in controlling microswimmer navigation and chemotaxis.

Main Methods:

• A computational model simulating microswimmers with shape adaptation controlled by artificial neural networks.
• Implementation of spatial and temporal sensing mechanisms for gradient detection.
• Utilizing the NeuroEvolution of Augmenting Topologies (NEAT) genetic algorithm for neural network evolution.

Main Results:

• Evolved simple neural networks effectively control microswimmer shape for navigation in static and dynamic chemical environments.
• Introduction of noise into the neural network successfully replicated the biased run-and-tumble motion observed in bacteria.
• Demonstrated that interpretable decision-making machinery coupled with environmental sensing facilitates navigation.

Conclusions:

• Artificial neural networks can evolve simple yet effective strategies for microswimmer navigation and chemotaxis.
• The model provides insights into the evolution of biological sensing and decision-making mechanisms.
• Findings are relevant for understanding intracellular sensing and simple nervous systems in organisms like *C. elegans*.