混合分支模型之蛋白質折疊問題模擬研究

Abstract

在計算生物學的研究上，預測蛋白質折疊後的結構以及功能，仍是現階段重要的課題。蛋白質的結構是由不同的胺基酸序列所構成，受到原子間及分子間的作用力，而互相吸引與排斥，發生摺疊。蛋白質的構造與蛋白質的功能有密切的相關性，若是蛋白質摺疊發生錯誤，導致形成不正常的構造，將會失去原有的功能與特性，引發疾病，在人類醫學的記錄上，如阿茲海默症(Alzheimer’s disease)與普昂疾病( Prion disease)，都是因為蛋白質發生錯誤的折疊，所引起的生理病變。 本論文將針對蛋白質摺疊預測問題，提出一套系統式的計算流程。在系統中，我們首先使用親疏水性晶格模型(Hydrophobic-Hydrophilic Lattice Model)，將胺基酸序列分成親水性與疏水性的單體，做直角與平角的折疊，並使用基因演算法(Genetic Algorithms)，來預測出蛋白質的初步立體結構。接著，我們引入親疏水性非晶格模型(Hydrophobic-Hydrophilic Off-Lattice Model)，將親疏水性晶格模型的運算結果，進行連續角度與扭轉度的運算，透過基因與禁忌搜尋演算法(Tabu Search Algorithm)，獲得蛋白質最為安定的結構，以及其結構所對應到的最小能量。我們結合晶格模型與非晶格模型，截取各模型的優點，我們可以降低運算的時間，並仍然保有好的精確度。最後，我們提出了分支模型(Branch Model )，這是一套新的模型。在分支模型中，我們將蛋白質長鏈中的肽鍵( Peptide )結構，視為極性分子團，並獨立考慮蛋白質單體中的側鏈( Side Chain )結構，由於側鏈的結構，將會決定蛋白質親疏水性的特性，進而決定蛋白質整體結構，以及蛋白質與蛋白質之間的作用關係，因此必須將側鏈所造成的影響，考慮到計算當中。 在分支模型的架構下，經由基因演算法與禁忌演算法，我們已經能更精確的預測出蛋白質的結構，對於預測蛋白質折疊後的藥物設計模擬程序，提供更高可靠性的資訊。

In the field of Computational Biology research, predicting protein structure and function after folding is a significant issue. Protein Structures are composed of different sequences of amino acid. Since interactions between atoms and molecules exist, attractive and repulsive forces are generated, and this results in folding. There is an affinity between protein structure and protein function. If the folding process fails, abnormal protein structures form, and the function and characteristics of protein are spent, which can lead to severe disease. Alzheimer’s disease and Parkinson’s disease are examples of the physiological changes caused by protein misfolding. This thesis offers a system-calculated process of protein folding prediction. In our calculated system, we first apply the Hydrophobic-Hydrophilic Lattice Model to classify amino acids into two types, characterized by hydrophobic and hydrophilic molecules. We then fold the protein sequence in perpendicular direction, or not fold, and apply the Genetic Algorithm to obtain the preliminary three-dimensional structure of the protein. Second, we apply the Hydrophobic-Hydrophilic Off-Lattice Model, and input the result of the first step into the system, considering the continuous bending angles and torsional angles in specific range, and then we apply the Genetic Algorithms and Tabu Search to get the stable protein structure and general minimum energy. Based on our calculated system, we propose that the Hybrid Model is composed of a Lattice Model and an Off-Lattice Model. We intercept the advantages of these two models, and not only reduce the computation time, but also retain good accuracy on energy calculation. Finally, we propose the Branch Model, which is the new model used in the protein folding simulation. In the Branch Model, we regard the peptide as the big polar molecules in the protein sequence, and take the side chains of each amino acid into account. Indeed, the characteristics of hydrophobic and hydrophilic amino acids, and the structure of protein after folding, even the protein-protein interaction, are related, in that the side chains of each amino acid have interactions with each other. This is the primary reason we propose the use of the Branch Model; we want to consider the influence of side chain interaction in protein folding calculation. The protein folding simulation results using the Branch Model that this thesis advances are more precise than either the Lattice Model or Off-Lattice Model. Moreover, we can reduce the computation time and retain good accuracy simultaneously. There is no doubt that we provide highly reliable information for drug design simulation after the step of predicting the protein folding structure.