Statistical Methods for Ultrahigh-dimensional Biomedical Data

描述：本提案开发了一种新的统计方法，从生物医学研究的高通量数据（如微阵列、蛋白质组学和下一代序列）中选择一小群分子，特别是用于自闭症研究和脑肿瘤。它专注于开发有效的方法和有效的统计工具，以控制错误发现率，测试一组分子的治疗效果，在变量误差、内质性和重尾误差分布的存在下进行特征选择和模型构建，以及预测临床结果和理解分子机制。它开发了半参数和非参数模型，以减少建模偏差和增加特征。它进一步发展了估计大协方差矩阵来理解遗传网络，统计模型的建立和推论。它引入了多元独立筛选和条件独立筛选技术，以减少变量筛选中的假阴性和假阳性，并为一系列统计模型开发了可计算和最优的惩罚似然方法。每种提出的方法的优缺点将通过理论调查和模拟研究进行批判性分析。开发相关软件。正在进行的自闭症研究、脑肿瘤和其他生物医学研究的数据集将使用新开发的方法进行分析，结果将得到进一步的生物学证实和调查。研究结果将对生物医学研究的高通量数据的统计分析以及对自闭症、脑肿瘤和其他疾病的分子机制的理解产生重大影响。

DESCRIPTION: This proposal develops novel statistical methods to select a small group of molecules from high-throughput data such as microarray, proteomic, and next generation sequence from biomedical research, especially for autism studies and brain tumors. It focuses on developing efficient methods and valid statistical tools for controlling false discovery rate an testing treatment effects on a group of molecules, for feature selection and model building in presence of errors-in-variables, endogeneity, and heavy-tail error distributions, and for predicting clinical outcomes and understanding molecular mechanisms. It develops semiparametric and nonparametric models to reduce modeling biases and to augment features. It furthers the developments on estimating large covariance matrices for understanding genetic network, statistical model building and inferences. It introduces multivariate independence screening and conditional independence screening techniques to reduce false negatives and false positives in variable screening, and develops computable and optimal penalized likelihood methods for an array of statistical models. The strength and weakness of each proposed method will be critically analyzed via theoretical investigations and simulation studies. Related software will be developed. Data sets from ongoing autism research, brain tumor, and other biomedical studies will be analyzed using the newly developed methods and the results will be further biologically confirmed and investigated. The research findings will have strong impact on statistical analysis of high throughput data for biomedical research and on understanding molecular mechanisms of autism, brain tumors, and other diseases.