Automated adaptive inference of phenomenological dynamical models

Area of Science:

• Complex Systems Science
• Computational Modeling
• Network Dynamics

Background:

• Complex systems involve intricate microscopic interactions, making them difficult to understand.
• Detailed models risk overfitting with limited data; simple models miss key features.
• Unknown parameters in dynamics hinder accurate mechanistic modeling.

Purpose of the Study:

• To develop a novel approach for constructing phenomenological, coarse-grained models of network dynamics.
• To create models that automatically adapt their complexity based on available experimental data.
• To ensure accurate predictions for complex systems even when microscopic details are unknown.

Main Methods:

• Developed an adaptive modeling approach for network dynamics.
• Constructed phenomenological, coarse-grained models.
• Ensured computational tractability for systems with many dynamical variables.

Main Results:

• The adaptive models accurately predict system behavior with limited data.
• Successfully inferred phase space structure for planetary motion using simulated data.
• Demonstrated avoidance of overfitting in a biological signaling system and accurate predictions for yeast glycolysis.

Conclusions:

• Adaptive, coarse-grained models offer a powerful approach to understanding complex systems dynamics.
• This method provides accurate predictions without requiring complete knowledge of microscopic details.
• The approach is computationally efficient and applicable to diverse scientific domains.