Modeling of nonlinear biological phenomena modeled by S-systems

Area of Science:

• Computational biology and systems biology
• Bioinformatics and computational genomics
• Biophysics and biochemical modeling

Background:

• Determining model parameters is a central challenge in computational modeling of biological systems.
• Dynamic genomic data can model genetic regulatory networks for disease intervention and understanding biological systems.
• Biological measurements are often noisy, requiring filtering to enhance their utility.

Purpose of the Study:

• To address state and parameter estimation for biological phenomena modeled by S-systems using Bayesian approaches.
• To compare the performance of various state estimation techniques, including Extended Kalman Filter (EKF), Unscented Kalman Filter (UKF), Particle Filter (PF), and Variational Bayesian Filter (VBF).
• To assess the impact of estimating multiple model parameters on estimation accuracy and convergence.

Main Methods:

• Utilized Bayesian approaches within a probabilistic state space model framework for nonlinear systems.
• Performed comparative studies on state estimation using noisy measurements of variables (e.g., enzyme CadA, cadaverine Cadav) in the Escherichia coli Cad System (CSEC).
• Evaluated simultaneous state and parameter estimation, analyzing accuracy and convergence with increasing numbers of estimated parameters.

Main Results:

• The Unscented Kalman Filter (UKF) demonstrated higher accuracy than the Extended Kalman Filter (EKF).
• The Variational Bayesian Filter (VBF) showed relative improvement over the Particle Filter (PF) by optimizing sampling distribution selection.
• Estimating more model parameters negatively impacted accuracy and convergence for all techniques, though VBF maintained advantages.

Conclusions:

• The Variational Bayesian Filter (VBF) offers superior accuracy and convergence for state and parameter estimation in noisy biological systems compared to EKF, UKF, and PF.
• The VBF's ability to optimize sampling distributions using observed data contributes to its enhanced performance.
• Careful consideration of the number of estimated parameters is necessary for robust biological model calibration.