Parameter Inference for an Astrocyte Model using Machine Learning Approaches.

Area of Science:

• Computational neuroscience
• Astrocytes and glial cell biology
• Machine learning in biological modeling

Background:

• Astrocytes are crucial glial cells supporting neuronal function and brain homeostasis, with dysfunction linked to neurological disorders like Alzheimer's and epilepsy.
• Accurate computational models of astrocytes are essential for research, but parameter inference remains a significant challenge.
• Existing computational models struggle with fast and precise parameter estimation.

Approach:

• Applied physics-informed neural networks (PINNs) to estimate parameters for a computational model of an astrocytic compartment.
• Incorporated dynamic weighting of loss components and Transformers to mitigate gradient pathologies in PINNs.
• Utilized a physics-informed neural network for control (PINC) adaptation to handle time-dependent input stimulations.

Key Points:

• Successfully inferred parameters for a computational astrocyte model using noisy, artificial data.
• Developed techniques to address gradient pathologies and time-dependent input challenges in PINNs.
• Achieved stable and accurate parameter estimation for astrocytic models.

Conclusions:

• Physics-informed neural networks, enhanced with specific techniques, offer a robust solution for parameter inference in complex astrocyte models.
• This methodology advances the development of computational neuroscience tools for studying brain function and disease.
• The stable parameter estimation facilitates better understanding and modeling of astrocyte roles in neurological conditions.