使用雙目標最佳化方法估算肌肉醣原分解之代謝路徑的動力學參數

Abstract

代謝路徑的模型重建是研究系統生物學的主要課題之一，建模的目的是為了研究蛋白質或整個代謝網路彼此之間的相互作用以及行為模式。例如肌肉之醣原分解是人體生理機能中不可或缺的一環，並且其代謝路徑的生化反應相當敏感。本研究的模型架構是以微分方程式為主軸的動力學模型，此模型能在反應過程中呈現出瞬間的濃度變化、酶所造成之通量及受酶影響之回應。 通常要重建一個動力學模型，例如肌肉醣原分解之代謝路徑模型，可使用被收集在特定資料庫中的數據，但可能有些參數值尚屬未知。這些動力學參數值通常是利用實驗方法去獲得，它們可能是從不同的實驗環境或是不同物種組成分批得來。由於代謝過程之中的動力學參數之間存在著熱力學限制以及反應代謝物的Gibbs生成自由能會決定反應平衡常數，使用這類參數值所建成的模型可能未遵守熱力學原理，導致不真實的生化反應。 本研究提出一套使用雙目標最佳化方法來估算代謝路徑模型的動力學參數並將之應用到肌肉醣原分解之代謝路徑的模型重建，其特色有(1)同時最小化系統總自由能來符合熱力學原理及最小化模型所估算反應物的濃度誤差及(2)使用智慧型基因演算法精確地解此雙目標最佳化問題的大量動力學參數。 從模擬實驗結果顯示最小化系統總自由能和最小化濃度誤差兩目標在提高模型精確度時有所衝突，使用雙目標最佳化方法有其必要。在模型精確度方面，我們以所提方法的估算參數值及現有方法的預設參數值所建的模型做比較，結果顯示我們的方法能獲得較低的系統總自由能(-0.72566kJ vs. -0.06561kJ)和較小的濃度誤差(提升準確率9.20%)。對所建模型做反應物與產物的敏感度分析，可進一步了解整個代謝網路彼此之間的相互作用。本文所提方法亦適用於其它相似代謝路徑的模型重建。

The model reconstruction of metabolic pathways is one of the major researches of systems biology, whose goal is to study the interaction and behaviors among proteins or whole metabolic networks. Muscle glycogenolysis is essential to the human physiological functions and its biochemical reactions of metabolic pathways are sensitive. In this study, the structure of kinetic models bases on the principle of ordinary differential equations, which can express the temporal changing, fluxes and responses influenced by enzymes in the progress of reactions. To construct a kinetic model, such as the metabolic pathways of muscle glycogenolysis, one can retrieve the values of kinetic parameters from specific databases. However, the parameter values for establishing the kinetic models were generally obtained from experimental methods, which may be derived separately from different experiments or combination of different species. Therefore, the constructed model may violate thermodynamics theory or be unknown, which results in unrealistic biochemical reactions. This study proposes a two-objective optimization approach to estimation of kinetic parameters for modeling metabolic pathways and its application to muscle glycogenolysis. The merits of this approach is twofold: 1) simultaneously minimizing the total Gibbs free energy of the system according to thermodynamic theory and minimizing the estimated concentration errors; and 2) using intelligent genetic algorithms to solve accurately the two-objective optimization problem with a large number of kinetic parameters. The simulation results reveal that the two objectives have conflicts in pursuit of accurate models. It means that the simultaneous optimization of the two objectives is necessary. In the aspect of model accuracy, the two models using the estimated and existing kinetic parameter values were compared. The results show that the proposed model has lower total Gibbs free energy of the system (-0.72566kJ vs. -0.06561kJ) and a smaller estimated concentration errors (improvement 9.20%). The sensitivity analysis on the reconstructed model can further understand the interactions of the metabolic networks. The proposed method is also effective to the model reconstruction of other similar metabolic pathways.