預測T細胞後天免疫反應

Abstract

發展電腦輔助疫苗設計系統能幫助免疫學家能快速有效的辨識候選疫苗，並且是免疫資訊學的終極目標之一。而精準的預測T細胞後天免疫反應是發展電腦輔助疫苗設計系統的關鍵。本研究之核心為發展能適用於探勘致免疫路徑（immunogenic pathways）中各種反應之重要物化特性的高性能大量參數最佳化演算法。此重要物理化學特性探勘系統之研發過程包含了三個重要的步驟：（1）蒐集各種能夠有效解釋生物現象之物理化學特性；（2）結合生物知識與演算法技巧來建立最佳化問題；（3）發展特定的高性能演算法來解決最佳化設計問題。重要特徵探勘系統可從大量資訊中探勘重要特徵來解釋各種免疫反應，並幫助建立T細胞後天免疫反應預測系統。 T細胞後天免疫反應包括有細胞毒性與輔助T細胞免疫反應。對於預測T細胞後天免疫反應，過去的研究多專注於建立主要組織相容性複合物（MHC）第一及第二型分子的抗原處理與表現路徑之預測模型。然而被主要組織相容性複合物結合的胜肽抗原並不一定能引起免疫反應。對於更複雜的T細胞免疫反應需要有更深入的研究並建立其預測模型。另外，對於抗原表現有重要影響的蛋白質泛素化（Ubiquitylation），至今仍未有預測模型。高度泛素化的蛋白因較容易被裂解，因而容易產生可供T細胞辨認用的抗原。因此準確的蛋白質泛素化預測將有助於辨識容易引起免疫反應的蛋白質抗原。 本研究專注在研究抗原的內生性物化特性，研發出第一套使用物化特性來預測與主要組織相容性複合物結合之蛋白質引起的T細胞免疫反應預測系統POPI與泛素化預測系統UbiPred。並發現過去普遍認同的抗原與主要組織相容性複合物的結合親和力並不足以準確預測T細胞免疫反應。針對影響細胞毒性與輔助T細胞免疫反應的重要物化特性之分析比較對於了解免疫反應有極大助益。本研究接著提出基於字串核函數的細胞毒性T細胞免疫反應預測模型POPISK。藉由融入主要組織相容性複合物與胺基酸位置的資訊，POPISK不僅能加強細胞毒性T細胞免疫反應之預測，同時也能準確預測由單一胺基酸突變引起的免疫反應變化。本研究並利用POPISK之特性來研究蛋白抗原上被T細胞辨認的重要位置。本篇研究結果將能幫助了解免疫系統並加速新疫苗的發展。

The development of computer-aided vaccine design systems is a goal of immunoinformatics that can largely accelerate the design of vaccines. Accurate prediction of adaptive T-cell immune response is the critical step to develop computer-aided vaccine design systems. The core of this study is to develop high-performance optimization algorithms for solving large-scale parameter optimization problems of bioinformatics to mine informative physicochemical properties from known experimental data for predicting immunogenic pathway. The development of these algorithms involves three major phases: (a) collection of physicochemical properties for encoding peptide sequences; (b) formulation of optimization problems using domain knowledge and computing techniques and, and (c) development of efficient optimization algorithms for solving optimization problems. The developed informative feature mining algorithms can be used to mine informative physicochemical properties for predicting peptide immunogenicity. There are two major T cells including cytotoxic and helper T cells. For the prediction of adaptive T-cell immune response, previous studies mainly focused on modeling antigen processing and presentation pathways of MHC class I and II. However, the prediction of T-cell response is much harder and less addressed because of the complex nature of T-cell response. Moreover, because over-ubiquitylated protein correlated with its half life, ubiquitylation plays an important role in providing antigen sources. Accurate prediction of ubiquitylation sites is helpful to identify immunogenic peptides. This study proposed the first prediction systems POPI and UbiPred for predicting T-cell response and ubiquitylation sites, respectively. The poor performance of a well recognized affinity-based method shows that binding affinity only is not sufficient for predicting T-cell response. The informative physicochemical properties for cytotoxic and helper T cells are identified and analyzed. Subsequently, an improved prediction system POPISK is proposed to predict cytotoxic T-cell response. The POPISK prediction system incorporating MHC allele information is used to identify important positions for T-cell recognition, and can predict immunogenicity changes made by single residue modifications. This study yields insights into the mechanism of immune response and can accelerate the development of vaccines.