A Novel Bayesian Model Averaging Approach for Genome Wide Association Studies

描述（由申请人提供）： 预防癌症的一个关键组成部分是揭示各种癌症背后的遗传学以及导致癌症的复杂特征和疾病。为了揭示癌症和其他复杂疾病或特征的遗传病因，有必要使用联合考虑疾病背后的多种遗传成分的方法。全基因组关联（GWA）研究使用扫描基因组的方法来寻找与疾病风险可能的遗传关联。然而，许多GWA研究使用单变量方法进行分析-将每个遗传标记视为独立的。最近，同时显著性检验和多变量层次模型的方法已经开始同时考虑多个基因，而不是单变量。当同时考虑标记时，这些方法将其自身限制于这样的假设，即当扫描基因组时，检测到的基因的数量与研究的基因的数量相比将非常小。作为回应，我们建议开发新的，更强大的工具，使用贝叶斯模型平均方法，包括模型中的遗传结构，同时在基因组范围内搜索复杂疾病（如肺癌）中的基因。包括生物信息的这种模型可以增加检测复杂疾病风险的小贡献者的能力，并且仍然可以包括控制假阳性的稀疏信息。最近，我们完成了一项方法学研究，表明贝叶斯模型平均比标准选择技术更好地使用多变量逻辑回归的假设驱动或候选基因类型的方法。该提案的中心主题是开发贝叶斯模型平均方法，该方法结合了GWA研究中使用的标记所固有的遗传结构，也可以搜索大量可用于GWA研究的标记。我们建议开发快速马尔可夫链蒙特卡罗算法的贝叶斯模型平均技术。我们将使用模拟研究校准新开发的统计技术，并应用新的和校准的方法，使用M。D.安德森癌症中心。该提案的意义在于开发新的GWA研究方法，这些方法将整合现有的生物信息，可以增加检测导致癌症的遗传因素的能力并控制假阳性。

DESCRIPTION (provided by applicant): A key component to preventing cancer is uncovering the genetics behind various cancers and the complex traits and diseases that lead to cancer. To uncover the genetic etiology for cancers and other complex diseases or traits, it is necessary to use methods that jointly consider multiple genetic components underlying the disease. Genome wide association (GWA) studies use methods to scan the genome looking for possible genetic associations with disease risk. However, many GWA studies perform the analysis using a univariate approach - treating each genetic marker as independent. Recently, methods for simultaneous significance testing and multivariate hierarchical models have started to consider multiple genes simultaneously, rather than univariately. While considering markers simultaneously, these methods restrict themselves to the assumption that when scanning the genome, the number of genes detected will be very small compared to the number of genes investigated. In response, we propose to develop novel, more powerful tools that use Bayesian model averaging methods to include genetic structure in the models, while simultaneously searching for genes in a complex disease, such as lung cancer, on a genome wide scale. Such models that include biological information can increase the power to detect small contributors to risk for complex diseases, and can still include sparsity information that controls for false positives. Recently, we completed a methodological study showing that Bayesian model averaging performs better than standard selection techniques using multivariate logistic regression in a hypothesis driven or candidate gene type approach. The central theme of this proposal is to develop Bayesian model averaging methods that incorporate genetic structure inherent to markers used in GWA studies that can also search through the immense number of markers available for GWA studies. We propose to develop fast Markov chain Monte Carlo algorithms for Bayesian model averaging techniques. We will calibrate the newly developed statistical techniques using simulation studies, and apply the new and calibrated methods to perform a GWA study of lung cancer using data already available at M. D. Anderson Cancer Center. The significance of this proposal is to develop new methods of performing GWA studies that will incorporate available biological information that can increase power and control false positives to detect genetic factors contributing to cancer.