辨識口蹄疫病毒抗原決定位之計算系統

Abstract

口蹄疫疾病主要影響偶蹄類動物是一個高度傳染性的疾病，被認為是世界上經濟方面最重要的家畜動物疾病。此疾病由口蹄疫病毒所造成，相關的研究認為疫情爆發時使用接種疫苗方式被視為一種更合理的手段。為了達到上述的目標，定義出具意義的抗原決定位及瞭解其重要的物化性質扮演一個重要的角色在保護性疫苗的設計、免疫的診斷及抗體的生產上。由於決定抗原上之B細胞表頂有賴於實驗的方式定義出來，但是此工作是耗時且昂貴的，因此發展計算的方式來幫助病原體自抗原的序列中定義出可靠性的表頂是必須地。 在此研究欲建立一個免疫的模型基於B細胞特異性的次分類群。我們擷取來自IEDB註解有關口蹄疫病毒病原體B細胞表頂之實驗數據做為我們的訓練資料。將訓練序列轉換成物化特性的特徵空間指標，並結合特徵選取的方式來改善預測準確度。使用繼承式雙目標基因演算法，其可最大化我們研究問題分類的準確度，同時最小化選取特徵數，來幫助擷取重要的資訊自我們的目標資料集之中。然後擴展所選出的物化性質組合來定義出抗原決定位的熱點藉由掃描病原體蛋白質。使用滑動窗口給予每一個詢問片段中心位點一個抗原性質倾向依據所選出的物化性質組合，結合投票的方式及利用智慧型基因演算法調整參數來達到一致性的預測結果。此外分析所選出物化性質來找尋重要的生物意義幫助改善疫苗的設計。 在這個研究中，我們發展一個計算的系統對於預測抗原決定位基於使用病原體特異性的次分類群及特徵選取的策略。結果顯示出，此建立的預測模型不僅能達到較高的預測準確度(訓練89.33%及測詴72%)，也能自病原體蛋白質序列中定義出抗原決定位的熱點。此外基於特徵選取也能提供有用的生物訊息供分析之用。此系統不僅可以被使用當做研究口蹄疫病毒新興病原體的工具而且可以提供一個概念對於改善B細胞表頂的預測上。

Foot-and-mouth disease (FMD) is a highly contagious disease affecting cloven-hoofed animals and it is deemed as economically important diseases of livestock worldwide. The causative agent is the foot-and-mouth disease virus (FMDV). The first priority of suggestion in the outbreak was to develop effective FMD vaccines. Accordingly, identifying significant antigenic determinant sites and understanding its important physicochemical properties play an important role in protective vaccine designs, immunodiagnostic tests and antibody production. The experimental methods for determining B-cell epitopes are time-consuming and expensive. Therefore, it is desirable to develop computational methods for reliable identification of putative B-cell epitopes from antigenic sequences. This study aims to establish a computational system for identifying antigenic determinant sites of foot-and-mouth diseases virus based on specific subclasses of B-cell epitopes of FMDV. We retrieved training data from the IEDB database and used the annotation of B-cell epitopes experimental data about FMDV. We transferred the training sequences to feature vectors based on the physicochemical feature index and then combined a feature selection method to improve prediction accuracy. An inheritable bi-objective genetic algorithm is used to maximize classification accuracy of the investigated problem and minimize the number of selected features to draw out significant information from our objective dataset. Then the selected feature set was spanned to identify hot points of the antigenic determinant sites by screening pathogen proteins. This method assigns a scale tendency value using the selected feature set and sliding windows of the query fragment. Moreover, we analyze the physicochemical feature set to mine significant biological findings to aid improve vaccine designs. The results showed that the prediction system could obtain high performance (training accuracy 89.33%, and test accuracy 72%) and identify promising putative antigenic hot points. Moreover, the feature selection method could provide much useful information for biological analysis. The prediction system is capable of identifying antigenic determinant sites from pathogen proteins. The system not only could be used as a tool for investigation of emerging pathogen strain of FMDV but also provides a conception to improve B-cell epitopes prediction effectiveness.