Enhanced segment particle swarm optimization for large-scale kinetic parameter estimation of escherichia coli network model / Mohammed Adam Kunna Azrag

Faculty of Computing

Thesis (Doctor of Philosophy) -- Universiti Malaysia Pahang – 2021

Includes bibliographical references

The development of a large-scale metabolic model of Escherichia coli (E. coli) is very crucial to identify the potential solution of industrially viable productions. However, the large-scale kinetic parameters estimation using optimization algorithms is still not applied to the main metabolic pathway of the E. coli model, and they’re a lack of accuracy result been reported for current parameters estimation using this approach. Thus, this research aimed to estimate large-scale kinetic parameters of the main metabolic pathway of the E. coli model. In this regard, a Local Sensitivity Analysis, Segment Particle Swarm Optimization (Se-PSO) algorithm, and the Enhanced Segment Particle Swarm Optimization (ESe-PSO) algorithm was adapted and proposed to estimate the parameters. Initially, PSO algorithm was adapted to find the globally optimal result based on unorganized particle movement in the search space toward the optimal solution. This development then introduces the Se-PSO algorithm in which the particles are segmented to find a local optimal solution at the beginning and later sought by the PSO algorithm. Additionally, the study proposed an Enhance Se-PSO algorithm to improve the linear value of inertia weight 𝜔 used in the Se-PSO. This algorithm added a damping process to increase the exploration and exploitation in the search space to support the particle to locate a global optimum solution. This modification facilitates an accurate determination of the optimal solution. The effectiveness of the adapted and proposed algorithms were evaluated using two experimental data (Chassagnole and Hoque) and statistically compared to the Particle Swarm Optimization algorithm (PSO), Genetic Algorithm (GA) and Differential Evolution (DE), based on the distance minimization, Mean, Standard Deviation (STD), and the F test (𝐹𝑡𝑒𝑠𝑡). The result of Se-PSO and ESe-PSO shows a tremendous impact on estimating the kinetic parameters where it can be inferred that distance minimization was achieved in all the algorithms. The ESe-PSO algorithm achieved (28.94% and 16.18%) distance minimization for Chassagnole and Hoque data as compared to the model under study (45.56% and 57.16%), Se-PSO (29.16%, 26.29%), PSO (29.36%, 37.09%), GA (35.58%, 33.9%), and DE (35.55%, 35.34%), respectively. Also, ESe-PSO achieved the best Mean (7.04E-05 and 7.41E-05) of the objective function compared to the Se-PSO algorithm best Mean (0.000603 and 0.00379), PSO (0.003893 and 0.00549), GA (0.11476 and 0.269007), and DE (0.049185 and 0.280478) respectively, for the 2 data set. Overall, the ESe-PSO and Se-PSO algorithms’ can be adopted effectively to estimate large-scale kinetic parameters to obtain accurate and acceptable results. Notably, the ESe-PSO superior to the original Se-PSO, PSO, and other state-of-the-art approaches in terms of distance minimization (accuracy), and the smallest objective function’s value produces appropriate fits to a two-set of experimental data.