Statistical methods for large-scale significance and prediction analysis with app

当前的技术进步为我们带来了大量的生物医学数据用于统计分析，例如癌症微阵列数据。这些数据的典型特征是，观察样本的数量远远少于变量/预测因子的数量，这给统计分析带来了挑战。识别差异表达基因和基于基因表达数据预测样本表型是分析这些大规模生物医学数据的两个重要研究问题。 该项目建议开发一些新的大规模预测和重要性分析统计方法，这些方法是专门为解决小样本量和潜在的样本异质性问题而设计的，结合现有的生物信息来改进推理，并且可以非常普遍地应用。这些方法的有效性将通过来源于白血病癌症研究项目的大规模生物医学数据来展示。癌症项目旨在通过识别用于关键早期治疗和快速非侵入性检测的分子生物标记物来改善癌症分子诊断和预后。 具体目标是：1)发展新的统计方法，用于大规模分子标记的显著性检验。2)发展新的统计方法，对样本异质性进行适当的建模，以进行显著性检验。3)发展新的统计方法，利用基因组信息来改进癌症预测。4)使用开发的模型和方法回答白血病癌症项目中与公共健康相关的研究问题；并在用户友好和有良好文档记录的软件中实施和验证所提出的方法，并免费分发给科学界。项目

Current technology advances have brought us massive biomedical data for statistical analysis, for example, the cancer microarray data. Typical of these data is the common feature that the number of observed samples is much smaller than the number of variables/predictors, which poses challenges for statistical analysis. Identifying differentially expressed genes and predicting sample phenotype based on the gene expressions data are two important research questions in analyzing these large-scale biomedical data. This project proposes to develop some new large-scale prediction and signifiance analysis statistical methods that are specially designed to address small sample size and potential sampe heterogeneity issues, incorporate existing biological information for improved inference, and can be applied very generally. The usefulness of these methods will be shown with the large-scale biomedical data originating from the leukemia cancer research projects. The cancer projects aimed to improve the cancer molecular diagnosis and prognosis by identifying molecular biomarkers for critical early treatment and rapid, noninvasive testing. The specific aims are 1) Develop new statistical methods for significance testing of large-scale molecular markers. 2) Develop new statistical methods that appropriately model the sample heterogeneity for significance testing. 3) Develop new statistical methods that utilize the gene group information to improve cancer prediction. 4) Use the developed models and methods to answer research questions relevant to public health in the leukemia cancer projects; and implement and validate the proposed methods in user-friendly and well-documented software, and distribute them to the scientific community at no charge. Project