Collaborative Research: Bayesian ANOVA for Microarrays

DNA微阵列可以洞察癌症等疾病阶段性发展过程中发生的基因变化。准确识别这些变化具有重要的治疗和诊断意义。然而，由于信息量巨大，对这些数据的统计分析是具有挑战性的。利用新的微阵列技术，可以测量每个被分析的组织样本近60,000个转录本上的表达。为了正确理解进展性疾病的演变，我们收集了所有可能的生物学阶段的表达值，因此，此类问题中的参数数量可以达到数十万，甚至数百万。高维性给两样本Z检验的标准ANOVA扩展带来了理论上的问题，Z检验是一种在两组中检测差异表达基因的流行方法。此外，专注于控制错误检测率的标准方法主要适用于更简单的实验设计；此外，这些方法往往是保守的，预计在多组设置下会更差。这项工作介绍了一种新的方法，称为用于微阵列的贝叶斯方差分析(BAM)，用于可靠地检测复杂实验环境中的差异表达基因。该方法依赖于高维变量选择方法，该方法利用了重新缩放的尖峰和板条分层模型。根据错误分类的基因总数，BAM被证明是风险最优的。这种风险最优化的确切机制在理论上被描绘为选择性收缩效应。理论指导自适应最优基因选择图形设备的开发。凯斯西储大学爱尔兰癌症中心收集的一个大型多阶段结肠癌微阵列资料库作为该方法的试验床。与之并行的是用于实现BAM的基于Java的软件的开发。该软件使用菜单驱动的图形用户界面，并包括最少数量的用户指定的调整参数，从而使其用户友好，可供其他分子生物学实验室使用。DNA微阵列允许对癌症等疾病的潜在遗传决定因素进行高通量分析。现在的典型情况是，每个被分析的组织样本都有近60,000个转录本上的表达。这些信息可以潜在地提供哪些基因参与癌症分期发展的信息，以及指示新的治疗和诊断靶点。然而，由于进行了大量的统计测试，识别感兴趣的基因的统计推断是具有挑战性的。标准方法使用ANOVA检验统计量，容易出现较高的错误检测。错误检测率控制方法往往过于保守，不会自然地扩展到更复杂的多阶段实验设计。这项工作介绍了一种新的方法，称为贝叶斯方差分析微阵列(BAM)，它可靠地检测差异表达的基因在多组实验设计设置。该方法采用了一种特殊的分层模型，它使基因选择具有神谕般的行为-也就是说，最终只有那些真正差异表达的基因被选择出来。这项研究从理论上描述了这种行为的原因，该理论指导了在真实微阵列数据集中自适应优化基因选择的新颖设备的开发。凯斯韦斯特恩储备大学爱尔兰癌症中心收集的一个大型多阶段结肠癌微阵列资料库作为方法的试验床，也为了解结肠癌的疾病过程提供了一个巨大的机会，这是一个非常重要的医学主题。虽然结肠癌根据临床分期有明确的演变，但对其分子进化知之甚少。与此并行的是使用菜单驱动的图形用户界面开发基于Java的软件，该图形用户界面具有最少的用户指定的调整参数，从而使将软件移植到分子生物实验室以主动用于分析其他疾病过程以及潜在的其他高通量数据源是可行的。

DNA microarrays can provide insight into genetic changes occurringduring stagewise progression of diseases like cancer. Accurateidentification of these changes has significant therapeutic anddiagnostic implications. Statistical analysis of such data is howeverchallenging due to the sheer volume of information. With newmicroarray technology it is possible to measure expressions on nearly60,000 transcripts for each sample of tissue analyzed. To properlyunderstand the evolution of a progressive disease, expression valuesare collected over all possible biological stages, thus the number ofparameters in such problems can be in the hundreds of thousands, oreven millions. The high dimensionality presents theoretical problemsto standard ANOVA-based extensions of two-sample Z-tests, a popularmethod for detecting differentially expressed genes in two groups.Additionally, standard approaches that focus on controlling falsedetection rates primarily apply to simpler experimental designs;moreover these approaches tend to be conservative and are expected tobe worse in multigroup settings. This work introduces a newmethodology called Bayesian ANOVA for Microarrays (BAM) for reliablydetecting differentially expressed genes in complex experimentalsettings. The method rests on a high dimensional variable selectionmethod that exploits a rescaled spike and slab hierarchical model.BAM is shown to be risk optimal in terms of the total number ofmisclassified genes. The exact mechanisms for this risk optimalityare theoretically delineated as a selective shrinkage effect. Theoryguides development of graphical devices for adaptive optimal geneselection. A large multistage colon cancer microarray repositorycollected at the Ireland Cancer Center of Case Western ReserveUniversity serves as a testbed for the methods. In parallel to thisis the development of JAVA-based software for implementing BAM.Software uses a menu driven GUI and includes a minimal number ofuser-specified tuning parameters, thus making it user friendly for useby other molecular biology laboratories.DNA microarrays allow for high throughput analysis of potentialgenetic determinants of diseases like cancer. It is now typical tohave expression on nearly 60,000 transcripts for each sample of tissueanalyzed. This information can potentially provide information aboutwhich genes are involved in stagewise development of cancer as well asindicate novel therapeutic and diagnostic targets. However,statistical inferences to identify interesting genes is challengingdue to the large number of statistical tests that are run. Standardapproaches employ ANOVA test statistics and are prone to high falsedetections. False detection rate control methods tend to be overlyconservative and do not extend naturally to more complex multistageexperimental designs. This work introduces a new methodology calledBayesian ANOVA for Microarrays (BAM) which reliably detectsdifferentially expressed genes in multigroup experimental designsettings. The method employs a special hierarchical model thatimparts an oracle like behaviour for gene selection --- that is,ultimately, only those truly differentially expressing genes areselected. The reasons for this behaviour are theoretically delineatedin this research, and the theory guides the development of novelgraphical devices for adaptively optimal gene selection in realmicroarray datasets. A large multistage colon cancer microarrayrepository collected at the Ireland Cancer Center of Case WesternReserve University serves as a testbed for the methods and alsoprovides a tremendous opportunity to understand the colon cancerdisease process, a topic which is of great medical importance. Whilecolon cancer has a well defined evolution defined by clinical stage,very little is known about its molecular evolution. In parallel tothis, is the development of JAVA-based software using a menu drivenGUI having a minimal number of user-specified tuning parameters, thusmaking it feasible to port the software to molecular biologylaboratories for active use in analysis of other disease processes andpotentially other high throughput sources of data.