應用Meta Decision Tree預測B細胞抗原表位

Abstract

抗原表位為抗原抗體的結合反應中抗原參與結合的部位，當B細胞感受器第一次與外部抗原接合時，B細胞將被激活並大量分裂增值，其中多數B細胞會分化成漿細胞並分泌大量抗體，這些抗體分子會接合目標抗原(外來物質)，將之中和或消滅。而剩餘部分B細胞則轉為記憶B細胞，當再次遇到相同抗原時，可立即識別並快速產生對應的抗體對抗該抗原。

準確的預測B細胞表位不僅可改善疫苗開發、疾病的診斷及治療，也可促進免疫學、醫療研究的發展。然而，使用傳統的理化實驗識別B細胞表位，其花費的時間以及成本皆相當的高。有鑑於此，越來越多更具成本效益的研究方法(如機器學習)被用於預測B細胞表位。

本篇研究主要目標在於延伸過去stacked和cascade等集成式學習方法，我們提出了Meta Decision Tree(MDT)以及3D球體結構鄰居特徵來幫助預測B細胞表位。為了證明新方法的性能，我們分析並比較當前具代表性的B細胞表位預測工具，而實驗結果證明我們的方法是有顯著改善B細胞表位的預測能力。

關鍵詞：B細胞表位、機器學習、表位預測。

B cell epitopes are part of the antigen involved in the antigen-antibody interaction. When a B cell is activated by its first encounter with an antigen that binds to its receptor, the cell proliferates and differentiates to generate a population of antibody-secreting plasma B cells and memory B cells. Most B cells differentiate into plasma B cells, will secrete large volumes of antibodies. These antibody molecules bind to the target antigen (foreign substance) and initiate its neutralization or destruction. The remaining portion of the B cells will become memory B cells. When faced with the same antigen again, they can immediately identify and rapidly produce antibodies against the corresponding antigen.

Accurate prediction of B cell epitopes can not only improve vaccine development, diagnosis and treatment of diseases, but also advance the development of immunology and medical research. However, conventional physicochemical experiments for identifying antigen epitopes are costly and time-consuming. In view of this, increasing efforts have been dedicated to the studies of more cost-effective computational methods for B-cell epitope prediction, using machine learning.

The aim of this thesis is to extend the idea of previous ensemble learning methods based on stacked and cascade generalizations by applying a meta decision tree approach and 3D sphere Neighborhood-based features. To demonstrate the performance of the new method, we conducted a thorough analysis and compared it with other state-of-the-art B-cell predictors, and the experimental results showed a marked improvement.

Keywords: B cell epitope, machine-learning, epitope prediction.