基因表达数据的双聚类分析与研究

发布时间：2018-03-15 20:00

本文选题：双聚类分析　切入点：基因表达数据　出处：《电子科技大学》2017年博士论文　论文类型：学位论文

【摘要】：如何快速从大规模基因表达数据中挖掘相关基因信息,实现高通量基因表达数据的精准分析,成为基因表达数据分析的关键问题。基因表达数据的双聚类分析能有效弥补传统聚类分析在搜索并确认基因局部表达模式的不足。本文以基因表达数据的双聚类分析为切入点,以提高双聚类体积、覆盖率、均方残差等质量评价指标和生物意义为主要目标,基于布谷鸟搜索算法,从单目标优化、多目标优化和集成学习等方面开展双聚类分析研究,解决现有双聚类分析方法存在的双聚类的质量差、多样性不足和生物意义不明显等问题。论文的主要工作包括:(1)提出基于布谷鸟搜索双聚类分析算法(Cuckoo Search Biclustering,CSB)。针对现有双聚类分析的低覆盖率和高均方残差等问题,该算法提出初始双聚类优化选取的策略提高解多样性,同时在搜索过程中采用服从莱维飞行的随机搜索策略解决解早熟。CSB算法可有效提高搜索范围和速度,并能稳定跳出局部最优解,同时可找到包含不同基因的双聚类,避免基因过于集中问题。与CC、FLOC、ISA、BIC-aiNet、SEBI、SAB和SSB等算法比较,实验表明CSB算法的双聚类质量和生物意义更优。(2)提出基于遗传算法和布谷鸟搜索的混合双聚类分析算法(Genetic Algorithm and Cuckoo Search hybrid Biclustering,GACSB)。通过引入遗传算法的锦标赛选择和精英保留等策略,GACSB算法可在计算代价不大幅增加的条件下拓展搜索范围和深度从而提高双聚类的多样性。与CC、FLOC、ISA、SEBI、SSB和CSB等算法的对比实验表明GACSB算法在双聚类的多样性和生物意义上有大幅提高。通过ACV、MSR和VE等指标对比分析,说明GACSB算法可搜索到不同类型的双聚类,具有较强可扩展性。(3)提出基于多目标布谷鸟搜索的双聚类分析算法(Multi-Objective Cuckoo Search Biclustering,MOCSB)。通过将双聚类分析转化为多目标优化问题,该算法把多目标布谷鸟搜索算法引入双聚类分析来同时优化双聚类的均方残差和体积等质量评价指标。MOCSB算法把搜索占优解集操作与布谷鸟巢搜索和宿主弃巢操作结合,可根据实际需要灵活使用各种双聚类评价指标。与CC、SEBI、SMOB和CSB等算法比较表明MOCSB算法能提高双聚类的质量和生物意义。(4)提出基于谱聚类的集成双聚类分析算法(Spectral Ensemble Biclustering,SEB)。针对双聚类集成问题中双聚类的质量不高且多样性不足,一致函数计算复杂度高和双聚类结果的生物意义不明显等问题,SEB算法使用不同双聚类质量评价指标获得多个基双聚类,然后基于谱聚类的一致函数进行集成获得一致双聚类。与VC、BGPC、MMMC和COAC等算法对比分析表明SEB算法在计算效率、双聚类的质量评价指标和生物意义等方面获得提高。
[Abstract]:How to quickly mine the related gene information from large-scale gene expression data, and realize accurate analysis of high-throughput gene expression data, It is the key problem of gene expression data analysis. The double cluster analysis of gene expression data can effectively make up for the deficiency of traditional clustering analysis in searching and confirming the local expression pattern of gene. In this paper, we take the double cluster analysis of gene expression data as the breakthrough point. In order to improve the quality evaluation indexes such as volume, coverage, mean square residuals and biological significance of the double clustering, based on the cuckoo search algorithm, double clustering analysis was carried out from the aspects of single objective optimization, multi-objective optimization and integrated learning. In order to solve the problem of poor quality of the existing methods of double clustering analysis, The main work of this paper includes: 1) to propose a Cuckoo Search Biclusteringing algorithm based on Cuckoo Search Biclustering.To solve the problems of low coverage and high mean square residuals in the existing double clustering analysis, In this algorithm, the initial double clustering optimization strategy is proposed to improve the diversity of solutions, and the random search strategy of flight from Levi to solve the precocious. CSB algorithm can effectively improve the search range and speed. It can stably jump out of the local optimal solution, at the same time, we can find the double clustering containing different genes, and avoid the problem of gene concentration, which is compared with the algorithms such as CCF FLOCU ISAA BIC-aiNet SEBISAB and SSB, etc. Experiments show that the CSB algorithm has better quality and biological meaning.) A hybrid clustering analysis algorithm based on genetic Algorithm and Cuckoo Search hybrid hybrid clustering is proposed based on genetic algorithm and Cuckoo search. The introduction of genetic algorithm for tournament selection and elite preservation is carried out. The GASCSB algorithm can expand the search range and depth without significantly increasing the computational cost, and thus improve the diversity of the double clustering. The experiment results show that the GACSB algorithm has diversity and growth in biclustering compared with other algorithms such as CCF FLOCU ISAA and SEBISSB and CSB. Through the comparative analysis of ACVG MSR and VE, It shows that the GACSB algorithm can search different types of double clustering, and has strong extensibility.) A multi-objective Cuckoo Search clustering analysis algorithm based on multi-objective cuckoo search is proposed. By transforming the double clustering analysis into a multi-objective optimization problem, this paper proposes a new algorithm, which is called Multi-Objective Cuckoo Search clustering algorithm. In this algorithm, the multi-objective cuckoo search algorithm is introduced into the double clustering analysis to optimize the quality evaluation indexes such as mean square residuals and volume simultaneously. MOCSB algorithm combines the search dominant solution set operation with the cuckoo nest search and host abandon nest operation. According to the actual needs, we can flexibly use all kinds of evaluation indexes of double clustering. Compared with the algorithms such as CCS, SMOB and CSB, it shows that the MOCSB algorithm can improve the quality and biological significance of the double clustering.) an integrated double clustering algorithm based on spectral clustering is proposed, which is called Spectral Ensemble Biclustering. In order to solve the problem of biclustering integration, the quality of biclustering is not high and the diversity is not enough. The problems of high computational complexity of uniform function and the biological significance of the result of biclustering are not obvious. The SEB algorithm uses different quality evaluation indexes of biclustering to obtain multiple base biclustering. Then, the coherent function based on spectral clustering is integrated to obtain the uniform biclustering, which is compared with the VCP-BGPC-MMMC and COAC algorithms. The results show that the efficiency of the SEB algorithm, the quality evaluation index and the biological significance of the BGPC-MMMC algorithm are improved.
【学位授予单位】：电子科技大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TP311.13

【相似文献】