基於蛋白質自由能之預測B細胞表位方法

Abstract

在預防醫學中，抗體可藉由疫苗刺激免疫系統產生，進而提升人體免疫力。胜肽疫苗，其抗原僅採用可誘發抗體之B細胞表位（B-cell epitope）胜肽片段刺激免疫系統就可以產生具特異性及保護性的免疫能力，因此有效的預測B細胞表位在預防醫學中扮演相當重要的角色。目前預測B細胞表位方法多數依賴由蛋白質結構所衍生出之胺基酸量表（amino acid propensity scales﹞。這類型的方法利用單一胺基酸序列做為預測B細胞表位之依據。然而大環境中病原不斷的演化，演化過程中病原的生物基因快速突變，基因突變也造成胺基酸序列發生改變。因此在病原快速演化的環境下，根據單一胺基酸序列所預測之B細胞表位做為疫苗的抗原並不一定適用。 在本研究中，我試著考慮一系列突變過後之胺基酸序列，並利用這一系列胺基酸序列所對應的蛋白質自由能設計出三種與自由能相關的特徵。利用這三種特徵配合 k-NN、 SVM 以及 ANN 這三種分類演算法，對於預測B細胞表位的預測分別可達到74.3%、66.1%及80.0%的準確性，與目前的B細胞表位預測方法 ── ABCPred、BCPred 和 AAP ── 相較，本研究所提出的方法可達到較好的預測效果。

Identification of B-cell epitopes plays an important role in vaccine development. Current prediction algorithms mostly rely on amino acid propensity scales and their variants, the results of which depend on a single antigenic phenotype. That viral sequences undergo continuous genetic changes, promoting the emergence of drug resistant strains, renders current prediction methods impractical. In this study, a novel set of features are proposed based on the protein free energy associated with point-mutated structures. To the best of our knowledge, this is the first attempt in this area to predict continuous B-cell epitopes based on protein free energy. I evaluated the novel features on k-nearest neighbor, support vector machine, and artificial neural network models, and achieved prediction accuracy of 74.3%, 66.1%, and 80.0% respectively. In comparison to current predictors, namely ABCPred, BCPred, and AAP, the energy-based models demonstrated better performance.