Code Generation in Computational Neuroscience: A Review of Tools and Techniques.

Area of Science:

• Computational Neuroscience
• Biophysics
• Scientific Computing

Background:

• Advances in experimental techniques and computational power drive complex computational neuroscience models.
• Large-scale biophysically detailed cell models and diverse point neuron models present significant computational challenges.
• Efficient and accurate transformation of mathematical model descriptions into executable code is crucial for all modeling methods.

Purpose of the Study:

• To provide an overview of existing code generation pipelines in computational neuroscience.
• To contrast the capabilities, technologies, and concepts behind these pipelines.
• To highlight the benefits of code generation for model description standardization and efficiency.

Main Methods:

• Review and analysis of existing code generation pipelines in computational neuroscience.
• Comparison of pipelines based on their aims, scope, and functionality.
• Discussion of the underlying technologies and concepts, including simulator-independent model description languages.

Main Results:

• Identified limitations of manual code translation (errors, inefficiency) and fixed model sets (limited flexibility).
• Highlighted the performance limitations of high-level interpreted languages for model definition.
• Demonstrated the growing popularity and advantages of code generation for automatic translation into efficient low-level code.

Conclusions:

• Code generation represents a promising third approach to address computational challenges in neuroscience modeling.
• This approach enhances standardization efforts for simulator-independent model descriptions.
• The reviewed pipelines offer diverse solutions, improving efficiency and accessibility in computational neuroscience.