Efficient microarray meta analysis and cancer biomarker selection

描述（由申请人提供）： 癌症研究的一个重要目标是鉴定可用于更好地理解癌症遗传基础的基因组生物标志物，并构建可用于预测癌症发生和进展的模型。许多研究已经使用微阵列来鉴定在各种癌症组织中改变表达水平的基因。Meta分析使得有可能（1）有效地将联合收割机实验与不同的微阵列平台和/或其他设置相结合;（2）导致跨研究的更可靠和一致的基因鉴定结果以及更令人满意的预测;以及（3）鉴定在不同类型的癌症中通常被激活的基因。 这项研究是第一个研究新的正规化方法的微阵列Meta分析，其中癌症的临床结果是测量沿着基因表达在多个独立的实验。所提出的方法可以（1）有效地联合收割机组合来自不同平台/实验设置的数据;（2）同时进行有效的生物标志物选择和预测模型构建;以及（3）识别在不同实验中重要的有影响力的基因，同时允许实验特异性预测模型。 本研究的具体目的包括：（1）建立用于基因芯片Meta分析的MTGDR（Meta Threshold Gradient Directed Regularization）方法。(2)开发规则化微阵列Meta分析的惩罚组桥方法。(3)将提出的一般方法应用于癌症分类和微阵列数据的生存分析。开发用户友好的R软件包，实现所提出的方法，并将其公开提供。 我们将考虑癌症微阵列Meta分析，其中单个实验可以有分类的临床结果和正确的删失生存结果。将对实际癌症研究和广泛的模拟进行分析，以评估所提出的方法的性能，并与替代方案进行比较。在这个应用中，我们强调的不仅是新的通用方法的发展，而且其计算机实现，应用和实证性能。

DESCRIPTION (provided by applicant): An important goal in cancer research is to identify genomic biomarkers that can be used to obtain a better understanding of the genetic basis of cancers, and construct models that can be used to predict cancer occurrence and progression. Many studies have used microarrays to identify genes that have altered expression levels in various cancer tissues. Meta analysis makes it possible to (1) effectively combine experiments with different microarray platforms and/or other setup; (2) lead to more reliable and consistent gene identification results across studies and more satisfactory predictions; and (3) identify genes that are commonly activated in different types of cancer. The proposed study is the first to investigate novel regularized methods for microarray meta analysis where cancer clinical outcomes are measured along with gene expressions in multiple independent experiments. The proposed approaches can (1) effectively combine data from different platforms/ experimental setup; (2) carry out efficient biomarker selection and predictive model building simultaneously; and (3) identify influential genes that are important across different experiments, while allowing for experiment-specific predictive models. The specific aims of this study include: (1) Develop MTGDR (Meta Threshold Gradient Directed Regularization) method for regularized microarray meta analysis. (2) Develop penalized group-bridge method for regularized microarray meta analysis. (3) Apply the proposed general methodologies to cancer classification and survival analysis with microarray data. Develop user-friendly R packages implementing the proposed approaches and make them publicly available. We will consider cancer microarray meta analysis where individual experiments can have categorical clinical outcomes and right censored survival outcomes. Analysis of practical cancer studies and extensive simulations will be conducted to assess performance of proposed approaches and compare with alternatives. In this application, we emphasize not only development of new general methodologies, but also their computer implementation, applications and empirical performances.