本文選題：微陣列數(shù)據(jù) + 統(tǒng)計推斷��； 參考：《第四軍醫(yī)大學》2009年碩士論文

【摘要】： 微陣列技術是生物技術變革的核心,允許研究者同時監(jiān)測成千上萬個基因的表達水平,已廣泛應用于醫(yī)學研究。如何挖掘海量基因表達信息中的有用信息,并進行生物學專業(yè)解釋是基因表達譜數(shù)據(jù)分析領域所面臨的一個重要挑戰(zhàn)。目前,針對海量基因表達數(shù)據(jù)不同學者和研究機構提供了各種統(tǒng)計分析方法和工具。本研究將這些方法大致劃分為兩大類:單基因分析(Single Gene Analysis,SGA)、基因集分析(Gene Set Analysis,GSA)。其目的都是為了能篩選出有差異表達的基因,以得到疾病的控制和預測。單基因分析不能有效地解釋生物學特性,且沒有考慮基因間的相關性,因此結論非常有限。自2003年Mootha等提出基因富集分析方法以來,微陣列數(shù)據(jù)基因集分析引起了統(tǒng)計學者與生物信息學者的廣泛關注。然而,由于基因表達譜數(shù)據(jù)本身特有的多維、樣本量小以及基因間復雜的相關性等特點,至今沒有一套成熟的理論和公認有效的篩選差異表達基因集的方法。本碩士課題結合實際微陣列數(shù)據(jù)、利用計算機技術和蒙特卡羅模擬研究微陣列數(shù)據(jù)基因集的統(tǒng)計分析理論方法及其應用,主要內(nèi)容包括基因集分析方法原假設的合理性、Ⅰ型錯誤的控制、篩選差異表達基因集(Different Expression Gene set,DEGs)的有效性等等。目前作了以下工作: 1.簡要介紹微陣列實驗基本概念、基因集注釋數(shù)據(jù)庫和單基因分析方法。在此基礎上廣泛復習和評價國內(nèi)外有關微陣列數(shù)據(jù)的基因集分析方法。按照基因集的定義、統(tǒng)計原假設框架與統(tǒng)計量理論分布的生成回顧和綜述了基因表達譜富集分析方法。 2.基因集分析原假設包括競爭性原假設(Q1)、自限性原假設(Q2)和混合型原假設(Q3)。更多的研究團體認為自限性原假設方法要好于基于競爭性原假設進行的統(tǒng)計推斷,但究竟哪種原假設更合理目前尚無定論。為了探討此問題,本研究通過模擬實驗進行比較研究。結果表明,自限性原假設方法檢驗效能較高,能識別出較多的差異表達基因集,但錯誤發(fā)現(xiàn)率較高;而競爭性原假設方法則是通過削弱其檢驗效能來達到較高的準確性;混合型原假設方法識別出的差異表達基因數(shù)及檢驗效能位于中間。我們建議進行微陣列數(shù)據(jù)分析時,如果條件允許可以采用不同原假設方法分析,否則采用混合型原假設,因為它克服了Q1、Q2方法的主要缺陷。 3.由于基因集統(tǒng)計量的概率密度函數(shù)未知,故一般采用重排列或有放回抽樣方法得到其理論分布。通常會認為重排列效果優(yōu)于反復抽樣,但是我們通過模擬實驗發(fā)現(xiàn)兩種效果基本一致,ROC曲線分析結果顯示有放回抽樣方法得到的曲線下面積稍大于重排列方法,說明同等條件下自助法抽樣略優(yōu)于樣本重排列。 4.假定基因間相互獨立的前提下,借助SAS 9.13模擬產(chǎn)生數(shù)據(jù)集,比較不同基因集方法篩選差異表達基因集的有效性。結果顯示Efron’s GSA方法的特異度及靈敏度均高于其它方法,而SAFE、Globaltest方法的檢驗效能僅次于Efron’s GSA方法。 5.由于基因間往往存在復雜的相關性,在模擬數(shù)據(jù)中納入這種相關關系。模擬實驗分析結果發(fā)現(xiàn)Efron’s GSA對此類數(shù)據(jù)完全失去判別能力,幾乎不能識別任何差異表達基因集。而PCOT2、Globaltest方法的效果卻非常顯著,能很好地識別模擬數(shù)據(jù)設定的差異表達基因集。 6.采用不同基因集方法對兩個著名的微陣列實驗數(shù)據(jù)進行實例分析比較。結論進一步證實考慮了基因間相關性基因集方法PCOT2、Globaltest優(yōu)于其他方法。而Globaltest方法能識別更多差異表達基因集,且模擬設定條件下錯誤發(fā)現(xiàn)率比PCOT2低19%。綜合模擬及實例數(shù)據(jù)分析結果,我們更傾向于主張采用模型分析法,如Globaltest方法(構建logistic隨機效應模型)進行基因集的分析。 本課題的創(chuàng)新點主要包括以下幾點:①針對不同原假設、理論分布生成方法對基因集分析結果的影響做了模擬比較研究。②將基因間相關性從不同角度納入模擬實驗數(shù)據(jù),分別模擬每個基因集內(nèi)部相關性,并基于此模擬數(shù)據(jù)進行基因集方法檢驗效能的比較。③模擬實驗結果顯示基于模型構建的基因集方法在數(shù)據(jù)分析時有效地考慮了基因間的相關性。④綜合實例比較后提出Globaltest是較有效的微陣列數(shù)據(jù)分析方法。 本課題主要是在微陣列數(shù)據(jù)基因集分析方法統(tǒng)計理論基礎上,對其所涉及的一些方法及相關問題進行了探索和研究,并給出了我們認為比較有效的基因表達譜數(shù)據(jù)分析法。期望能夠為陜西省科技計劃攻關項目(微陣列數(shù)據(jù)差異表達信息挖掘及應用研究,編號:2008K04-02)的下一步研究工作打下良好基礎,為基因表達微陣列數(shù)據(jù)的統(tǒng)計分析方法,尤其是基因集分析提供參考。
[Abstract]:microarray technology is the core of biotechnology change , which allows researchers to monitor the expression level of thousands of genes at the same time and has been widely used in medical research . The aim is to screen out the differentially expressed genes in order to get the control and prediction of diseases . The single gene analysis can not effectively explain the biological characteristics , and therefore , it is very limited . Since the gene expression profiling data itself has many characteristics such as multi - dimensional , small sample size and complex correlation between genes , the main contents include the rationality of the original hypothesis of gene set analysis method , the control of type I error , the selection of different expression gene set ( DEGs ) , and so on . The following work is currently done :



1 . The basic concepts of microarray experiments , gene set annotation database and single gene analysis method are introduced briefly . On this basis , we review and evaluate the gene set analysis method about microarray data at home and abroad . According to the definition of gene set , the generation of statistical original hypothesis framework and statistical quantity theory distribution is reviewed and summarized . The analysis method of gene expression profiling is summarized .



2 . The original hypothesis of gene set analysis includes competitive source hypothesis ( Q1 ) , self - limiting original hypothesis ( Q2 ) and mixed primary hypothesis ( Q3 ) . More research groups believe that self - limiting original hypothesis method is better than statistical inference based on competitive original hypothesis .



3 . Because the probability density function of gene set statistics is unknown , the theoretical distribution is usually obtained by the method of re - arranging or sampling . It is usually considered that the re - arrangement effect is superior to the repeated sampling , but we find that the two effects are basically consistent through the simulation experiment , and the ROC curve analysis results show that the area under the curve obtained by the sampling method is slightly larger than that of the re - sampling method , so that the self - service method sampling under the same condition is slightly better than the sample rearranger .



4 . On the premise of mutual independence of genes , the data set was simulated by SAS 9.13 , and the validity of different gene set methods was compared . The results showed that the specificity and sensitivity of the method were higher than those of other methods .



5 . Because of the complex correlation among genes , this correlation was included in the simulation data . The results of the simulation experiment showed that the Eron ' s gsa completely lost the discrimination ability to such data , almost unable to identify any differentially expressed gene sets . The results of the PCOT2 and Globaltest methods were very significant , and the difference expression gene set of the simulated data set can be well recognized .



6 . Two well - known microarray experimental data were analyzed and compared with different gene sets . The conclusion further confirmed that the gene set method PCOT2 and Globaltest were better than other methods . The Globaltest method could identify more differentially expressed gene sets , and the error rate was 19 % lower than that of PCOT2 under simulated set conditions .



The innovation points of this project mainly include the following points : ( 1 ) the influence of the theory distribution generation method on the results of gene set analysis is simulated and compared for different original hypotheses . 鈶，

本文編號：1801933