Improving dynamic predictions with ensembles of observable models

Area of Science:

• Systems biology
• Computational biology
• Mathematical modeling

Background:

• Dynamic mechanistic modeling in systems biology faces challenges due to complex interactions and limited experimental data.
• Ensemble modeling, adapted from statistical mechanics, uses multiple models to describe biological phenomena but can be unreliable for predicting unobservable states due to identifiability issues.

Purpose of the Study:

• To develop and validate a strategy for assessing and enhancing the reliability of model ensembles in systems biology.
• To improve the predictive power of kinetic models, particularly for internal biological dynamics.

Main Methods:

• Utilized a global optimization metaheuristic for parameter estimation to build diverse model ensembles.
• Integrated structural identifiability and observability analysis to detect and address potential prediction issues.
• Developed model reparameterization techniques to overcome identified prediction limitations.

Main Results:

• Successfully created ensembles of kinetic models (using ordinary differential equations) with improved reliability.
• Quantified uncertainty in state trajectory predictions.
• Demonstrated the approach's efficacy on diverse biological models including glucose regulation, cell division, circadian rhythms, and JAK-STAT signaling.

Conclusions:

• The proposed strategy enhances the reliability of model ensembles for predicting internal biological dynamics.
• This method provides a robust framework for systems biology modeling, addressing identifiability and observability challenges.
• The developed methodology and code are publicly available for broader application.