细菌IV型分泌系统效应蛋白的特征挖掘、计算预测及初步实验鉴定研究

项目名称： 细菌IV型分泌系统效应蛋白的特征挖掘、计算预测及初步实验鉴定研究

项目编号： No.31301097

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 生物科学

项目作者： 邹凌云

作者单位： 中国人民解放军第三军医大学

项目金额： 18万元

中文摘要： 巴尔通体等病原菌具备IV型分泌系统(T4SS)，可转运效应蛋白至宿主细胞内，逃避宿主的防御并维持细菌生存。鉴定T4SS效应蛋白对于揭示细菌的致病机制极为重要。当前的鉴定方法以实验为主，在速度和成本上都不能满足在大量病原菌基因组内进行研究的需要，而同时也缺乏计算预测的手段来解决这一矛盾。本课题组在前期研究中，计算氨基酸位置特异性特征，构建支持向量机模型，预测T4SS效应蛋白准确度达到91.9%，表明建立普适性预测方法是可行的。本研究将在此基础上，挖掘效应蛋白的多类特征，筛选特征子集，采用机器学习算法融合策略建立预测模型，采用数据集交叉验证和文献验证评估预测性能，进一步提高预测精度并开发web预测程序。此外，以汉赛巴尔通体为对象，建立分层分类模型，在其基因组内预测候选效应蛋白，采用Cya-fusion实验进行证实，期望鉴定1到多个新的效应蛋白分子，揭示该菌通过T4SS与宿主相互作用的新途径。

中文关键词： IV型分泌系统；效应分子；特征挖掘；机器学习；在线预测服务器

英文摘要： Type IV secretion systems (T4SS) exist in many pathogens such as Bartonella, which transport various effectors into host cells to evade the defense of hosts and increase bacterial survival. Identification of T4SS effectors is extremely important to reveal the mechanisms of bacterial pathogenicity. Most of existing methods are experiment-based, which are not cost-effective for fast T4SS effector identification in a large number of pathogen genomes. The absence of accurate bioinformatics prediction methods is keeping this problem unsolved. In our previous studies, we have established an effector dataset and computed position specific features of sequence residues to train a support vector machine, which can predict T4SS effectors with the accuracy of 91.9%.The result shows that universal prediction of T4SS effectors is feasible.On this basis, we will mine multi-class sequence features using computational approaches and screen feature subset using feature selection methods. An ensemble prediction model based on machine learning algorithms will be constructed. The performance of our model will be tested in datasets via cross-validation tests, as well as by literature authentication. The overall accuracy of prediction is expected to be improved, and a web server will be developed for prediction applications. In addit

英文关键词： type IV secretion systems；effectors；feature mining；machine learning；web server