RateML: A Code Generation Tool for Brain Network Models.

Area of Science:

• Computational Neuroscience
• Neuroinformatics
• Scientific Computing

Background:

• Whole brain network models are crucial for research but pose significant informatics challenges.
• Existing workflows lack efficient simulation implementation for complex models.

Purpose of the Study:

• Introduce RateML, a tool to simplify the generation of whole brain network models.
• Enable efficient simulation implementation across different hardware, including GPUs.

Main Methods:

• RateML utilizes NeuroML's Low Entropy Model Specification (LEMS) to describe model mathematics declaratively.
• It generates simulation code for CPUs and GPUs (CUDA C++ for NVIDIA GPUs).
• RateML facilitates parameter tuning by enabling parallel ensemble simulations.

Main Results:

• RateML significantly reduces parameter exploration time for brain network models.
• Generated CUDA code allows exploration of thousands of parameter combinations on a single GPU.
• The tool accelerates research workflows, enabling broader parameter fitting and larger cohort studies.

Conclusions:

• RateML offers a new approach to create efficient brain network models and simulation workflows.
• It supports more robust and statistically relevant conclusions about brain dynamics.
• The tool democratizes high-performance computing for neuroscience research.