Computational Prediction of MHC Class II Epitopes

描述(申请人提供)：新的T细胞表位的发现将极大地促进改进疫苗的设计和开发，因为它为选择主要组织相容性复合体(MHC)分子和可诱导T细胞激活的抗原肽之间的复合体提供了关键信息。表位识别的关键步骤之一是预测MHC与多肽的结合。两类主要的MHC分子参与了两种类型T细胞表位的产生。由于表位的结合基序相对保守，预测MHC I类表位的方法已经取得了相对较高的准确性。然而，由于表位的长度不定，每个表位的核心区不确定，以及未知氨基酸作为主锚，MHC II类表位的预测方法的性能受到了阻碍。大多数现有的方法试图通过各种比对技术来识别一组表位的结合核心。然后，可以从识别的比对中组装用于预测的结合基序或位置特定评分矩阵。在文本挖掘技术的启发下，我们开发了一个用于MHC II类表位预测的监督学习模型的原型。其思想是通过迭代过程从由表位和非表位组成的训练集中区分核心结合二聚体和非核心二聚体。该模型的特点是简单，能够同时利用表位和非表位信息。初步研究表明，该模型对人类白细胞抗原-DR4(Bl*0401)表位具有良好的预测性能。在这项研究中，我们计划对该模型进行彻底的评估和优化。在目标1中，我们将制定模型的优化原则，并选择方法的最佳变种。在目标2中，我们将针对现有的各种等位基因特异性数据的主要预测因素进行彻底的评估。最后，在目标3中，我们将建立一个网络服务器，用于预测各种MHC II类等位基因特异性表位。该系统将向研究界免费提供。我们的长期目标是开发预测T细胞表位的计算方法。计算预测可以为含有免疫刺激序列的病原体分子提供一种快速方法，这些序列可以作为免疫干预或诊断的靶标。

DESCRIPTION (provided by applicant): The discovery of novel T cell epitopes will greatly facilitate the design and development of improved vaccines by providing critical information needed for the selection of complexes between the major histocompatibility complex (MHC) molecules and antigen peptides that can induce T cell activation. One of the key steps for the epitope identification is the prediction of MHC-peptide binding. Two major classes of MHC molecules are involved in the generation of two types of T cell epitopes. Methods for the prediction of MHC class I epitopes have achieved relatively high accuracy, since the binding motifs of the epitopes are relatively conserved. However, the performance of the prediction methods for MHC class II epitopes are hindered by the variable lengths of the epitopes, the undetermined core region for each individual epitope, and the unknown amino acids as primary anchors. Most of the existing methods attempt to identify binding cores for a set of epitopes through various alignment techniques. Binding motifs or the position specific scoring matrices for prediction can then be assembled from the identified alignment. Motivated by a text mining technique, we have developed a prototype of an supervised learning model for the MHC class II epitope prediction. The idea is to discriminate the core binding nonamers from the non-core nonamers derived from a training set consisting of epitopes and non-epitopes through an iterative process. The characteristics of this model are the simplicity and the capacity of using information both from epitopes and non-epitopes. The preliminary study demonstrated promising performance of this model for HLA-DR4 (Bl*0401) epitopes. In this study, we plan to conduct a thorough evaluation and the optimization of this model. In Aim 1, we will develop the principle for optimization of the model and select the best variant of the method. In Aim 2, we will conduct a thorough evaluation against existing major predictors for various allele specific data. Finally, in Aim 3, we will establish a web server for the prediction of various MHC class II allele-specific epitopes. The system will be freely available to the research community. Our long-term goal is the development of computation methods for prediction of T-cell epitopes. The computational prediction can provide a rapid method for the of pathogen molecules containing immunostimulatory sequences that can serve as targets for immune intervention or diagnostics.