Model discrimination in dynamic molecular systems: application to parotid de-differentiation network.

Area of Science:

• Systems Biology
• Computational Biology
• Biostatistics

Background:

• Longitudinal data analysis, such as tracking mRNA concentrations, is crucial in systems biology.
• Ordinary differential equations (ODE) models are common but may lack flexibility for sparse molecular data.
• Model discrimination is essential for identifying the most suitable model under data scarcity.

Purpose of the Study:

• To propose a formal, sensitive, and flexible method for discriminating between competing Bayesian mixture-type longitudinal models.
• To address the challenges of model selection with limited data points in molecular biology.
• To develop a Bayes discriminant rule applicable to complex biological variability.

Main Methods:

• Application of concepts from Bayesian analysis of computer model validation.
• Utilization of modern Markov Chain Monte Carlo (MCMC) algorithms.
• Development of a Bayes discriminant rule for two-model comparison problems.

Main Results:

• A formal method for discriminating between Bayesian mixture-type longitudinal models was derived.
• The method demonstrated sensitivity and flexibility for complex longitudinal molecular data.
• The rule was successfully applied to mammalian salivary gland mRNA data and a synthetic dataset.

Conclusions:

• The proposed Bayes discriminant rule offers a robust approach for model selection in sparse longitudinal data scenarios.
• This method enhances the reliability of systems biology models by formally falsifying inappropriate candidates.
• The approach is valuable for analyzing complex molecular data, such as mRNA concentrations in biological networks.