Optimizing Biophysical Large-Scale Brain Circuit Models With Deep Neural Networks.

Area of Science:

• Computational Neuroscience
• Systems Neuroscience
• Biophysics

Background:

• Biophysical models offer mechanistic insights into brain function at various scales.
• Optimizing model parameters is crucial for biological plausibility but computationally demanding.
• Existing optimization methods struggle with scalability due to repeated numerical integration.

Purpose of the Study:

• To introduce a novel deep learning framework, DELSSOME, for efficient biophysical model parameter optimization.
• To bypass computationally expensive numerical integration in biophysical modeling.
• To accelerate large-scale mechanistic modeling in neuroscience.

Main Methods:

• Developed DELSSOME (DEep Learning for Surrogate Statistics Optimization in MEan field modeling).
• Framework predicts realistic brain dynamics directly from model parameters, avoiding numerical integration.
• Integrated DELSSOME into an evolutionary optimization strategy.

Main Results:

• DELSSOME achieved a 2000× speedup over Euler integration for the FIC model.
• Trained DELSSOME models generalized to new datasets without retraining.
• Enabled a 50× speedup in FIC model estimation while maintaining neurobiological insights.

Conclusions:

• DELSSOME significantly accelerates biophysical model optimization, overcoming scalability limitations.
• The framework facilitates large-scale mechanistic modeling in population neuroscience.
• This acceleration unlocks new avenues for understanding complex brain function.