基于蛋白质网络的人类遗传致病基因预测算法研究

发布时间：2018-11-14 18:19

【摘要】： 人类遗传疾病基因的预测是目前生物信息学研究中的一个热点问题。随着基因组测序的完成和新一代测序技术的发展,基因以及蛋白质相互作用网络数据不断增长。通过这些数据来分析预测致病基因给人类破译疾病的遗传基础和分子基础带来了新的方法,对基因组学和医学都具有重要的现实意义。本文依据遗传疾病临床描述与蛋白质相互作用网络的关联关系,获取相关数据,首先对人类遗传疾病在线数据库进行文本挖掘,通过向量空间模型的方法,计算遗传疾病表型之间的重叠关系,其次利用蛋白质相互作用关系数据计算相互作用评分,加上疾病-蛋白质关联关系数据,共同构建了一系列生物网络,对疾病表型和蛋白质网络数据进行分析,基于这些网络数据对疾病的候选致病基因进行打分和排序,来预测疾病基因。本文提出了两种新的基于网络的人类遗传疾病基因预测方法。第一种在传统的二维关联概率模型的基础上,提出了基于多维随机变量的两种新的概率模型——中心概率模型和外形概率模型,在此基础上,提出了基于多概率融合的致病基因预测数学模型。还有一种是使用过滤函数的回归分析预测,通过观察网络和概率模型数据,总结数据规律,加入生物学假设,使用过滤函数融合回归分析进行候选基因的排序。基于多概率融合的预测模型通过综合利用联合概率、中心概率和外形概率,在多个层面上反映了表型相似性与蛋白质相互作用之间的关联规律,与传统的基于单一概率的模型相比,该数学模型更好地反映了生物系统的真实面貌,具有更强的疾病基因预测能力。而另一个使用过滤函数融合回归分析的预测模型,根据合理的生物学假设,反映疾病表型重叠中蕴含的基因关联关系,可以有效的将很多非致病基因排在队列的后面,有效增强了预测方法的效率。
[Abstract]:Gene prediction of human genetic diseases is a hot issue in bioinformatics. With the completion of genome sequencing and the development of new generation sequencing technology, the data of gene and protein interaction network are increasing. These data are used to analyze and predict the genetic basis and molecular basis of human disease, which is of great practical significance to genomics and medicine. In this paper, according to the relationship between clinical description of genetic diseases and protein interaction network, we obtain relevant data. Firstly, we mine the online database of human genetic diseases by vector space model. The overlapping relationship between phenotypes of genetic diseases was calculated, and then a series of biological networks were constructed by using protein interaction relation data to calculate interaction score and disease protein correlation data to construct a series of biological networks. The disease phenotypes and protein network data were analyzed and the candidate disease genes were ranked and sequenced based on these network data to predict disease genes. In this paper, two new genetic prediction methods for human genetic diseases based on network are proposed. First, based on the traditional two-dimensional correlation probability model, two new probabilistic models based on multidimensional random variables, the central probabilistic model and the shape probabilistic model, are proposed. A mathematical model for predicting pathogenic genes based on multi-probability fusion is proposed. The other is regression analysis with filter function. By observing the data of network and probability model, summarizing the rule of data, adding biological hypothesis, the candidate gene is sorted by filtering function fusion regression analysis. The prediction model based on multi-probability fusion reflects the correlation law between phenotypic similarity and protein interaction on many levels by making comprehensive use of joint probability, center probability and shape probability. Compared with the traditional model based on single probability, the mathematical model can better reflect the real appearance of the biological system and has a stronger ability to predict disease genes. Another predictive model using filter function fusion regression analysis, based on reasonable biological assumptions, reflects the genetic association in disease phenotypic overlap, and can effectively rank many non-pathogenic genes at the back of the queue. The efficiency of the prediction method is improved effectively.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2010
【分类号】：R394

【相似文献】