Efficient simulation of neural development using shared memory parallelization.

Area of Science:

• Computational neuroscience
• Developmental biology
• Software engineering

Background:

• Simulating complex neural development is computationally intensive.
• Existing models may lack efficient parallel processing capabilities.
• Understanding neural morphological growth, migration, and pruning is key to developmental neuroscience.

Purpose of the Study:

• To introduce NeuroDevSim, a Python module for simulating key aspects of neural development.
• To present an agent-based modeling approach for neural growth, migration, and pruning.
• To demonstrate efficient parallel processing for large-scale neural simulations.

Main Methods:

• Utilizes an agent-based modeling approach with 'fronts' for neural development processes.
• Implements a novel shared memory approach for multi-core parallel processing without messaging.
• Employs a coding rule to avoid memory conflicts, with a serialized lock broker for collision detection.

Main Results:

• NeuroDevSim simulates morphological growth, migration, and pruning of neurons.
• Achieves linear strong parallel scaling up to 96 cores, successfully running on 128 cores.
• Enables modeling of diverse neural networks, from few complex to thousands of simple neurons.

Conclusions:

• NeuroDevSim provides an efficient and scalable platform for simulating neural development.
• The shared memory approach offers significant parallel processing advantages.
• Facilitates research in computational neuroscience and developmental biology through accessible modeling.