Sensitivity, robustness, and identifiability in stochastic chemical kinetics models.

Area of Science:

• Computational Biology
• Chemical Kinetics
• Systems Biology

Background:

• Stochastic chemical kinetics models are crucial for understanding cellular processes.
• Calculating Fisher information is essential for model analysis but computationally intensive.
• Existing methods often rely on computationally expensive Monte Carlo simulations.

Purpose of the Study:

• To develop a novel, efficient numerical method for computing Fisher information matrices in stochastic chemical kinetics.
• To enable robust analysis of model sensitivity, robustness, and parameter identifiability.
• To provide a computational tool for designing experiments in cellular stochastic processes.

Main Methods:

• Utilized the linear noise approximation to derive model equations and a likelihood function.
• Developed an efficient algorithm that reduces Fisher information matrix calculation to solving ordinary differential equations.
• Implemented the algorithm as a Matlab package.

Main Results:

• Presented the first method to compute Fisher information for stochastic chemical kinetics without Monte Carlo simulations.
• Demonstrated significant differences between stochastic and deterministic models, and between time-series and time-point measurements in stochastic models.
• Identified molecular number variability, species correlations, and temporal correlations as sources of discrepancies.

Conclusions:

• The novel method offers an efficient alternative for Fisher information calculation in stochastic chemical kinetics.
• Discrepancies between stochastic and deterministic models highlight the importance of considering molecular noise.
• The approach facilitates the analysis and design of experiments for probing cellular stochastic processes.