【作者】 吴娟；

【导师】 骆嘉伟；

【作者基本信息】 湖南大学 ， 信息与通信工程， 2014， 硕士

【摘要】 作为生命活动中必不可缺的部分,关键蛋白质有助于人们理解有机体的新陈代谢、分化等生命过程。关键蛋白质的识别从系统层面上为生物学的发展和医药学研究提供重要价值信息。后基因组时代,蛋白质相互作用数据日益丰富,从蛋白质网络层面识别关键蛋白质成为研究的热点。然而,由于蛋白质相互作用数据不全,并且存在假阳性噪声数据,关键蛋白质识别的准确率受到影响,如何提高关键蛋白质识别的准确率成为关键蛋白质研究的热点问题。针对蛋白质相互作用网络中较多的假阳性噪声数据和缺失的数据限制了拓扑中心性关键蛋白质识别算法的精度这一问题,本文利用多维生物特征识别关键蛋白质,通过评估蛋白质的生物功能重要程度,减小网络中数据缺失和假阳性数据带来的负面影响,从而提高关键蛋白质识别的准确性。关键蛋白质常常密集分布于某些功能的蛋白质复合物内,而在其他的复合物内较少出现。并且,共表达程度高的蛋白质复合物进化更缓慢,这与关键蛋白质不谋而合,关键蛋白质的保守性与共表达程度高的蛋白质复合物存在紧密的联系。本文提出CED算法,该方法基于共表达复合物、基因表达程度等生物特征和边聚集系数拓扑特征识别关键蛋白质。我们将CED算法和现有的几种具有代表性的算法应用于酵母蛋白质真实网络上,实验结果表明,CED算法正确识别的关键蛋白质数量一致高于其他几种算法。为进一步提高关键蛋白质识别的精确度,本文深入探索能更直接有效地表征蛋白质功能重要程度的生物特征。网络模体内的蛋白质的功能和进化程度更相似,并且模体也是进化保守的。本文利用模体、基因表达数据和蛋白质复合物等多维生物特征识别关键蛋白质,提出MGC算法。将MGC算法应用于DIP和MIPS酵母蛋白质真实网络上。采用统计指标分析、jackknife验证法、Top100关键蛋白质的差异集分析等方法验证MGC算法的有效性。实验结果表明,MGC的性能优于其他几种算法。更多还原

【Abstract】 Essential proteins help people understand life processes of organisms,such as metabolism,differentiation and so on.They are indispensable for life activities.Essential proteins provide valuable information for the development of biology and medical research from the system level.In post genome era,protein-protein interactions data increase exponentially,and this opens a window for identifying essential proteins from protein interaction network.However,the interactions in the protein interaction networks are still incomplete and contain false positive data.The accuracy of topological centrality based algorithms is deeply affected by this noise data and missing data.Therefore,it is a challenge for improving the prediction precision of essential proteins.To reduce the negative effects of missing data and false positive data in the network,multi-dimensional biometrics are used to identify essential proteins in this paper.We mainly base on the functional importance of proteins to identify essential proteins.To some extent,we reduced the dependence on the network and improved the prediction precision.Essential proteins are said to densely distribute in some protein complexes.Protein complexes with high co-expression level often evolve slowly,and this is coinciding with essential proteins.Essential proteins are closely related to highly co-expressed protein complexes.Therefore,CED proposed in this paper is based on co-expression level of protein complex,gene expression level and edge clustering co-efficient.It is the combination of biological features and topological property.CED and several existing algorithms are applied on yeast protein network.Experimental results show that CED identifies more essential proteins than other algorithms both on DIP network and BioGRID network.In order to improve the prediction precision of essential proteins further,biology functional features which represent proteins more directly and efficiently are needed to be further explored.We observed that proteins in the same motif tend to show similar functions and evolve at a similar rate,and motif is said to evolve conserved.Therefore,a new algorithm for predicting essential proteins based on motif,gene expression and protein complex is proposed.It is named after MGC.MGC and six previously proposed algorithms are applied on yeast protein interaction networks which are downloaded from DIP database and MIPS database.Jackknife methodology,F-measure et al.are the means used to validate the effectiveness of MGC.Experimental results show that the performance of MGC is better than other six algorithms.更多还原