Developing Explainable Deep Model for Discovering Novel Control Mechanism of Neuro-Dynamics

Area of Science:

• Computational neuroscience
• Systems biology
• Artificial intelligence

Background:

• Understanding the human brain's complex dynamics requires robust computational models, often using partial differential equations (PDEs).
• Empirical model tuning based on domain knowledge limits discovering novel mechanisms from spatiotemporal data.
• A gap exists in linking data-driven discovery with mechanistic understanding of brain functions.

Purpose of the Study:

• To develop an explainable deep model integrating deep learning and physics principles for analyzing the human brain.
• To uncover the mechanistic role of the brain in maintaining controllable functions under external stimulation.
• To investigate the latent control mechanisms of neurobiological processes at a system level.

Main Methods:

• A unified framework combining deep neural networks with physics principles was designed.
• Optimal control theory guided the development of explainable deep models for neurobiological processes.
• The model was applied to uncover the pathophysiological mechanisms of Alzheimer's disease.

Main Results:

• The developed model provides a system-level understanding of neurobiological dynamics.
• Novel insights into the pathophysiological mechanisms of Alzheimer's disease were uncovered.
• The model demonstrated improved prediction accuracy and explainability for disease progression compared to black-box models.

Conclusions:

• Explainable deep models offer a powerful approach to bridge data-driven insights and mechanistic understanding in neuroscience.
• This framework enhances the ability to study complex systems like the human brain and diseases such as Alzheimer's.
• The findings pave the way for more accurate disease progression prediction and better understanding of disease etiology.