Optimal experiment design for model selection in biochemical networks.

Area of Science:

• Biochemistry
• Computational Biology
• Systems Biology

Background:

• Mathematical modeling formalizes hypotheses in biochemical networks, aiding model discrimination.
• Bayesian model selection quantifies evidence for competing models, favoring simpler ones.
• Insufficient experimental data often hinders clear model distinction.

Purpose of the Study:

• Develop a novel method for Optimal Experiment Design (OED) to guide model selection.
• Predict which experiments will most effectively discriminate between competing biochemical network models.
• Enhance the ability to select the best model when data is scarce.

Main Methods:

• Applied a Bayesian approach to infer model parameter distributions.
• Utilized sampling from multivariate predictive densities for simulation.
• Employed k-Nearest Neighbor estimation of Jensen Shannon divergence between predictive densities.

Main Results:

• Demonstrated successful model selection using predictive differences in test cases.
• Showcased the method's flexibility across various model quantities.
• Identified specific experimental combinations that improve model discriminability.

Conclusions:

• The OED method effectively enables model selection through predictive differences.
• The approach is highly flexible due to its basis in predictive distributions.
• It integrates with existing Bayesian methodologies for enhanced posterior analysis.