Parametric sensitivity analysis for biochemical reaction networks based on pathwise information theory.

Area of Science:

• Biochemistry
• Computational Biology
• Systems Biology

Background:

• Stochastic modeling and simulation are crucial for understanding complex biochemical networks.
• Parameter calibration in these models is challenging due to high dimensionality.
• Existing sensitivity analysis methods face computational limitations with many parameters.

Purpose of the Study:

• To develop an efficient sensitivity analysis methodology for stochastic reaction networks.
• To address the computational cost associated with high-dimensional parameter spaces.
• To improve the understanding of parameter robustness and identifiability in biochemical models.

Main Methods:

• Developed a pathwise sensitivity analysis based on Information Theory.
• Quantified information loss due to parameter perturbations using time-series distributions.
• Employed the Relative Entropy Rate and defined a pathwise Fisher Information Matrix (FIM).

Main Results:

• The pathwise approach offers a gradient-free method for parameter sensitivity.
• The FIM reveals hidden parameter dependencies and sensitivities within reaction networks.
• The block-diagonal structure of the FIM aids in analyzing parameter relationships.

Conclusions:

• The proposed method significantly benefits the analysis of complex stochastic systems.
• FIM structure aids in efficiently addressing parameter identifiability, estimation, and robustness.
• Validated on diverse biochemical models including protein production, p53 network, and EGFR signaling.